Professional Documents
Culture Documents
THE PAST DECADE HAS seen an unusual pattern of investment. The boom
of the 1990s generated unusually high investment rates, particularly in
equipment, and the bust of the 2000s witnessed an unusually large decline
in investment. A drop in equipment investment normally accounts for
about 10 to 20 percent of the decline in GDP during a recession; in the
2001 recession, however, it accounted for 120 percent.
1
In the public mind, the recent boom and bust in investment are directly
linked due to capital overhang. Although the term is not very precisely
dened, this view generally holds that excess investment in the 1990s,
fueled by an asset price bubble, left corporations with excess capital
stocks, and therefore no demand for investment, during the 2000s. The
popular view also holds that these conditions will continue until normal
economic growth eliminates the overhang and, consequently, that there is
little policymakers can do to remedy the situation, by subsidizing invest-
ment with tax policy, for example. Variants on this view have been
espoused by private sector analysts and economists,
2
and the notion of a
285
MI HI R A. DES AI
Harvard University
AUS TAN D. GOOL S BEE
University of Chicago
We thank Mark Veblen and James Zeitler for their invaluable research assistance, as
well as Alan Auerbach, Kevin Hassett, John Leahy, Joel Slemrod, and participants at the
Brookings Panel conference for their comments. Dale Jorgenson was kind enough to pro-
vide estimates of the tax term by asset. Mihir Desai thanks the Division of Research at Har-
vard Business School for nancial support. Austan Goolsbee thanks the American Bar
Foundation and the National Science Foundation for nancial support.
1. McCarthy (2003) documents the decline in equipment investment as a share of GDP
for all business cycles since 1953 and shows the 2001 recession to be an outlier.
2. See, for example, Berner (2001); Graeme Leach, The Worries of the World,
GCIEye no. 1, 2002, accessed August 2004 (www.gcieurope.com/eye/GCIEye_Issue01.
pdf), and Stephen Roach, The Costs of Bursting Bubbles, New York Times, September 22,
2002, section 4, p. 13.
2581-04_Desai.qxd 1/18/05 13:28 Page 285
capital overhang has certainly been on the minds of leading Federal
Reserve ofcials and researchers.
3
Whether or not a capital overhang is the true explanation of the
investment bust, it is clear that the drop in investment has motivated
policymakers to try to stimulate investment through ambitious fiscal
policy changes.
4
Under President George W. Bush, depreciation allow-
ances for equipment investment have been increased twice, in 2002 and
2003, and in 2003 the tax rate on dividend income was cut sharply and
that on capital gains income more modestly. These measures were mainly
intended to increase after-tax returns and stimulate investment. The typi-
cal analysis of the investment collapse and policy response is summarized
by the Republican chairman of the Joint Economic Committee:
Excessive and bad business investments made during the stock market bubble
have taken years to liquidate. In nine of the 10 quarters beginning with the
fourth quarter of 2000, real business investment has declined. Fortunately,
recent tax legislation signed into law in 2003 should promote business invest-
ment by increasing the after-tax returns from investing in capital assets and
alleviating nancing constraints among small and medium-size rms.
5
Yet, after several years of tax cuts, investment has still not risen impres-
sively compared with previous recoveries. This contrast has reignited
claims that tax policy is ineffective at stimulating investment, although
some make the more specic charge that tax policy is impotent when it
follows a period of excessive investment.
This paper examines the evidence on the two related issues of capital
overhang and taxes using data at the industry, the asset, and especially
the rm level. Specically, we address two questions: rst, did over-
investment in the 1990s cause the low investment of the 2000s, and, second,
did investment in the 2000s become less sensitive to prices, and does this
explain why tax policies, specically the equipment expensing and the
dividend tax cuts of 2002 and 2003, seem to have been ineffective in
restoring investment to normal levels?
286 Brookings Papers on Economic Activity, 2:2004
3. See, for example, Greenspan (2002), Ferguson (2001), Bernanke (2003), French,
Klier, and Oppedahl (2002), Pelgrin, Schich, and de Serres (2002), Kliesen (2003), Doms
(2004), and McCarthy (2004).
4. Unlike the behavior of investment, the behavior of tax policy in the 2000s is com-
pletely consistent with earlier time periods. Cummins, Hassett, and Hubbard (1994) have
documented that a primary determinant of investment tax subsidies is a drop in investment.
5. Saxton (2003).
2581-04_Desai.qxd 1/18/05 13:28 Page 286
We begin by examining the degree to which growth in investment
during the boom was correlated with a decline in investment during the
bust across different assets and industries. There are, of course, many
possible definitions of overhang or excess investment. We will not try to
show that there was no overoptimism in product or capital markets.
Clearly equity prices rose substantially and then fell, as did investment
rates. Instead we investigate whether investment grew the most in those
assets and industries in which it subsequently declined the most. We
want to know if any aftereffects of the investment boom of the 1990s
persisted into the 2000swhether firms behaved differently because too
much capital remained from the investment decisions of the 1990s.
The evidence across assets, industries, and firms suggests that, con-
trary to the popular view, there is little correlation between the invest-
ment boom of the 1990s and the investment bust of the 2000s. We also
present some more specific evidence, using firm-level data, that invest-
ment behavior has remained just as responsive to the fundamentals (as
measured by Tobins q) regardless of how much a firms investment
grew or how much its market value rose in the 1990s. Essentially, we
find that the explanatory power of the standard empirical model of
investment has not deteriorated in the 2000s, despite the common per-
ception that it has.
We then use that standard model to consider the impact of tax cuts. To
estimate the impact of the dividend tax reduction, we revisit an enduring
debate in public nance between the new view of dividend taxation,
which says that dividend tax cuts do not reduce the cost of capital for mar-
ginal investments, and the traditional view, which says that such cuts do
reduce the marginal cost of capital and thus stimulate investment. The
evidence at the rm level strongly supports the new view and suggests
that the dividend tax reductions enacted in 2003 had little or no effect on
investment.
Finally, to estimate the impact of the changes in depreciation allow-
ances, we estimate a tax-adjusted q model similar to that of Lawrence
Summers,
6
but with greater emphasis on the importance of error in the
measurement of q, as emphasized by Jason Cummins, Kevin Hassett, and
Mihir A. Desai and Austan D. Goolsbee 287
6. Summers (1981).
2581-04_Desai.qxd 1/18/05 13:28 Page 287
288 Brookings Papers on Economic Activity, 2:2004
7. Cummins, Hassett, and Hubbard (1994).
8. Tevlin and Whelan (2003) argue that much of the increase in gross investment can
be explained empirically by falling prices of computers and their higher depreciation rates.
Glenn Hubbard.
7
The method introduced for handling these measurement
error issues yields results that suggest that both tax policy and q are likely
to have much larger effects on investment than found in the traditional lit-
erature, where coefcients are very small and imply implausibly large
costs of adjustment. Even with the more reasonable adjustment costs, how-
ever, we show that the depreciation allowance changes of 2002 and 2003
changed the tax term by a relatively small amount: the estimated overall
impact in these two years was an increase in investment of only 1 to 2 per-
cent, far too small to offset the double-digit declines of the early 2000s.
Capital Overhang and Investment
Real investment was considerably higher than normal during the late
1990s. When recession years are excluded, investment from 1947:1 to
1995:2 averaged about 12 percent of GDP; the highest quarterly level was
15 percent in 1984:3. From 1996:1 to 2000:4, in contrast, this ratio aver-
aged more than 16 percent, and it reached 18 percent at its peak. The dis-
tinctiveness of these investment rates holds even relative to the business
cycle. Figure 1 shows that investment in the quarters leading up to the
2001 peak was higher than it had been during comparable periods in pre-
vious cycles. The popular view holds that this extra investment resulted
from the excesses of the 1990s bubble.
8
With this view in mind, gure 2 provides a counterpart to gure 1,
showing the path of investment in the period after the trough quarter for
the recovery that began in late 2001 and for the average of previous
recoveries. The increase in investment in the current recovery, at least
through the beginning of 2004, is notably smaller than in the average
recovery. Taken together, these trends make it plausible to many observers
that investment after the most recent trough was lower than in previous
cycles precisely because investment in previous years had been higher.
Of course, these aggregate patterns do not establish any underlying
connection between the rise and the fall. To test for a causal relationship,
we believe, it is critical to disaggregate the investment data. Most aca-
2581-04_Desai.qxd 1/18/05 13:28 Page 288
Mihir A. Desai and Austan D. Goolsbee 289
demic work looking at the capital overhang has not done so, or has done so
at a very broad level, emphasizing that the reversal in investment has been
concentrated in information technology investment.
9
It is clear that the
exuberance of the 1990s was not shared equally in all sectors. Industries
such as telecommunications experienced huge increases in the 1990s in a
way that railroads or mining, say, did not. We believe that any overhang,
in the sense of excess capital remaining at the end of the boom, is inher-
ently an industry- or rm-level phenomenon, which requires that we look
at data at that level.
An additional reason to look at the industry- and rm-level data is that
investment theory typically begins with the premise that there is a per-
fectly functioning secondary market for capital goods and a at supply
curve for capital. In such a world, rms with an overhang of unused cap-
ital equipment can simply sell it without incurring any loss. For the
9. Two papers by McCarthy (2001, 2004) are exceptions.
0.8
0.9
1.0
12 8 4
Quarters before peak
Source: Authors calculations using National Income and Product Accounts data.
a. Investment is measured as a fraction of GDP.
Average of previous postwar cycles
19962000
Index = 1.0 at peak
Figure 1. Real Investment Relative to Business Cycle Peak, Postwar Average and
19962000
a
2581-04_Desai.qxd 1/18/05 13:28 Page 289
popular view to make sense, then, either investment must be in some way
irreversible (which leads to a rather different model),
10
or there must be
some other type of adjustment cost associated with disinvestment.
11
Matthew Shapiro and Valerie Ramey have documented that, in some
industries, a sizable wedge can develop between the purchase price of
capital goods and the sale price.
12
These types of irreversibilities are
likely to be rm or asset specic rather than applying to all types of invest-
ment in all sectors homogeneously. Fortunately, data on investment are
available at the industry, asset type, and rm level, and, as we will show,
the evidence at all three levels of disaggregation is generally the same.
290 Brookings Papers on Economic Activity, 2:2004
10. As in, for example, Abel and Eberly (2002).
11. The adjustment costs could be at the rm level, or they might be external in the
sense that the supply of capital goods in a particular industry is upward sloping as in Gools-
bee (1998, 2001).
12. Ramey and Shapiro (2001). Evidence presented by Goolsbee and Gross (2000) is
also consistent with that view.
1.0
1.1
1.2
1.3
4 8 12
Quarters past trough
Source: Authors calculations using National Income and Product Accounts data.
a. Investment is measured as a fraction of GDP.
200104
Averages of previous postwar cycles
Index = 1.0 at trough
Figure 2. Real Investment Relative to Business Cycle Trough, Postwar Average and
200104
a
2581-04_Desai.qxd 1/18/05 13:28 Page 290
Mihir A. Desai and Austan D. Goolsbee 291
13. Before 1997, Standard Industrial Classication (SIC) codes were employed.
14. See, for example, Auerbach and Hassett (1992), who discuss the problems with
estimating structures investment.
Evidence at the Industry Level
We begin with the evidence on changes in investment at the industry
level. Rather than rely on the more aggregated xed asset data available
from the Bureau of Economic Analysis (BEA), we turn to the Annual
Capital Expenditure Survey (ACES) of the U.S. Census Bureau, which pro-
vides a ner industry disaggregation than is available elsewhere. The sur-
vey samples approximately 46,000 companies in more than 100 industries,
categorized according to the 1997 North American Industry Classication
System (NAICS). We narrow this categorization down to eighty-one
nonoverlapping industries at approximately the three-digit NAICS level.
13
The ACES provides measures of gross investment only and does not
estimate industries capital stock. Consequently, we cannot scale invest-
ment by lagged capital as is done in traditional empirical work on invest-
ment. Instead we simply investigate the change in total investment both for
equipment alone and for equipment and structures combined. Empirical
models of investment have struggled to explain the behavior of investment
in structures, and it is not known whether this problem is due to mismea-
surement in the tax term, unobservable factors in structures markets (such
as liquidity and nancing issues relating to the supply side of the market),
or some other factor.
14
Since we cannot readily isolate equipment invest-
ment from structures investment in the rm-level data employed below,
we have to assume that equipment investment and overall investment
behave in the same way. Given that, by the 2000s, equipment accounted
for something like 80 percent of nonresidential investment, this may not
be too heroic an assumption, but the results in these areas will allow us to
check the results in a circumstance where we have both sets of data.
Our goal with these data is to look for general evidence supporting the
view that a capital overhang from the 1990s is a key factor determining
investment in the 2000s. If overhang is quantitatively important, one
might expect to nd that those industries in which investment has fallen in
the 2000s are the same as those in which investment grew substantially in
2581-04_Desai.qxd 1/18/05 13:28 Page 291
the 1990s. To test this relationship formally, we performed a cross-
sectional regression of the change in log investment by industry from
2000 to 2002 (the period widely viewed as the collapse) on the change
in log investment from 1994 to 1999 in the same industry, estimating the
following equation:
(1) ln(I
i, 2002
) ln(I
i, 2000
) + [ln(I
i, 1999
) ln(I
i, 1994
)] +
i
.
This test would show no evidence of reversion, of course, if all indus-
tries boomed and then busted together equally, since any such effect
would simply appear in the constant term. Given that the investment
growth of the 1990s was not likely to have been identical across indus-
tries, this equation provides a useful estimation strategy.
Table 1 presents, in the top panel, the results of estimating equation 1
by ordinary least squares (OLS) and, in the bottom panel, results for the
same specications employing median regressions, to ensure that the
292 Brookings Papers on Economic Activity, 2:2004
Table 1. Regressions of Changes in Investment during 200002 on Changes in
Investment during the 1990s, Equipment Only, Using Data by Industry
a
Manufacturing Manufacturing
All industries only All industries only
Sample period 1-1 1-2 1-3 1-4
Ordinary least squares regressions
199499 0.084 0.5315
(0.0693) (0.1894)
199799 0.0582 0.4204
(0.0853) (0.2047)
199497 0.1435 0.7878
(0.1331) (0.2704)
No. of observations 81 23 81 23
Adjusted R
2
.018 .273 .022 .330
Median regressions
199499 0.0205 0.5836
(0.0994) (0.2210)
199799 0.0239 0.5712
(0.1392) (0.2753)
199497 0.1164 0.8117
(0.2206) (0.4492)
No. of observations 81 23 81 23
Source: Authors regressions using data from the Annual Capital Expenditure Survey of the U.S. Census Bureau.
a. The dependent variable in all regressions is the change in log capital expenditure on equipment from 2000 to 2002; the inde-
pendent variable is the same change in log values in the same industry for the indicated sample period. Numbers in parentheses
are standard errors.
2581-04_Desai.qxd 1/18/05 13:28 Page 292
results in the top panel do not purely reect the role of large outliers. Col-
umn 1-1 reports the results from the basic overhang specication. The
OLS and median regressions produce coefcients that are negative but
very small and not signicantly different from zero. To give a sense of the
magnitude of this effect, a 1-standard-deviation change in the investment
rate from 1994 to 1999 (0.53, or from the median of 0.38 to about the 85th
percentile) is associated with only a 2.9 percent lower level of investment
(less than one-twelfth of a standard deviation) from 2000 to 2002. This is
modest evidence of an overhang, at best.
Given the serious decline of manufacturing in the most recent reces-
sion, and given that old-line manufacturing was not typically involved in
the Internet boom, we further investigate the manufacturing sector sepa-
rately. In column 1-2, which restricts the sample to the twenty-three
manufacturing industries, the evidence for an overhang seems more pro-
nounced. In both the OLS and the median regressions, there is a large and
signicant negative coefcient on the change in investment from 1994 to
1999. In the median regression, a 1-standard-deviation increase in the
investment rate among manufacturing industries (a log value of 0.32) in
199499 corresponds to an almost 22 percent lower investment rate in
200002, which is equal to about two-thirds of the standard deviation of
those changes. If one takes this larger effect as evidence of overhang (as
opposed to a cyclical phenomenon), however, it should be noted that
manufacturing industries accounted for only about 22 percent of total
equipment investment and 18 percent of total investment in 2002, accord-
ing to the ACES.
15
Consequently, on the present evidence, mean reversion
for manufacturing can explain only a limited part of the aggregate col-
lapse of investment.
The common explanation for capital overhang is that an abundance of
funds raised in the capital market during the bubble encouraged the
excess investment, particularly during the 199799 period. Indeed, the
broadly disaggregated, cost of capitaltype analysis done by Jonathan
McCarthy suggests that there was no capital overhang at all until 1998,
even in the high-technology investment goods sector (computers and
communications equipment).
16
In columns 1-3 and 1-4, therefore, we con-
Mihir A. Desai and Austan D. Goolsbee 293
15. This is also consistent with the evidence cited by Bernanke (2003).
16. McCarthy (2003).
2581-04_Desai.qxd 1/18/05 13:28 Page 293
sider separately the periods from 1994 to 1997 and from 1997 to 1999 in
order to isolate the effects of the bubble period and to take account of
underlying growth trends in different industries that might mask invest-
ment reversion. Again there is little evidence of reversion across all
industries, and there are larger negative coefcients in manufacturing.
The later period, in which the overhang is alleged to have occurred, has a
smaller coefcient than the earlier period, although the standard errors are
not small enough to reject the hypothesis that they are equal. Rather than
supporting the intuition of a bubble-induced capital overhang, this consid-
eration of the two subperiods suggests some underlying, more secular
mechanism associated with the continuing decline in U.S. manufacturing.
Table 2 considers the behavior of both equipment and structures
investment. The results are qualitatively similar to those in table 1 in that
they show little evidence of reversion, either generally or in manufactur-
ing, featuring the dynamics discussed earlier.
294 Brookings Papers on Economic Activity, 2:2004
Table 2. Regressions of Changes in Investment during 200002 on Changes in
Investment during the 1990s, Equipment and Structures, Using Data by Industry
a
Manufacturing Manufacturing
All industries only All industries only
Sample period 2-1 2-2 2-3 2-4
Ordinary least squares regressions
199499 0.0516 0.5426
(0.0645) (0.169)
199799 0.0677 0.4786
(0.0871) (0.1916)
199497 0.0304 0.6663
(0.1001) (0.2395)
No. of observations 81 23 81 23
Adjusted R
2
0.008 0.329 0.009 0.247
Median regressions
199499 0.0533 0.6793
(0.1066) (0.2030)
199799 0.0285 0.5450
(0.1182) (0.3532)
199497 0.1186 0.6564
(0.1184) (0.3665)
No. of observations 81 23 81 23
Source: Authors regressions using data from the Annual Capital Expenditure Survey of the U.S. Census Bureau.
a. The dependent variable in all regressions is the change in log capital expenditure on equipment and structures from 2000 to
2002; the independent variable is the change in log values in the same industry for the indicated sample period. Numbers in
parentheses are standard errors.
2581-04_Desai.qxd 1/18/05 13:28 Page 294
Evidence at the Asset Level
Next we consider the general evidence on investment by type of invest-
ment good rather than by industry. We did this by testing our basic regres-
sion model (equation 1) using data by asset category instead of by
industry. Using only the BEA data available for the whole period
19942002, we have twenty-ve different categories of equipment and an
additional nine categories of structures.
17
As in tables 1 and 2, the two
panels of table 3 report both OLS and median regressions. (Weighting by
the initial capital stock in these regressions provides very similar results,
which we do not report here.) Those asset types that had the largest
increases in investment from 1994 to 1999 are not systematically those
that had the largest drop in investment from 2000 to 2002: the regressions
Mihir A. Desai and Austan D. Goolsbee 295
17. The categories of structures employed by the BEA change slightly over the period.
Table 3. Regressions of Changes in Investment during 200002 on Changes in
Investment during the 1990s, Using Data by Asset Type
a
Sample period 3-1 3-2
Ordinary least squares regressions
199499 0.0945
(0.0970)
199799 0.0388
(0.1576)
199497 0.2326
(0.1611)
No. of observations
b
34 34
Adjusted R
2
.029 .063
Median regressions
199499 0.1490
(0.1105)
199799 0.0545
(0.1270)
199497 0.2407
(0.1113)
No. of observations
b
34 34
Source: Authors regressions using National Income and Product Accounts data from the Bureau of Economic Analysis.
a. The dependent variable in all regressions is the change in log capital expenditure from 2000 to 2002; the independent vari-
able is the same change in log values for the same asset type for the indicated sample period. Numbers in parentheses are standard
errors.
b. Sample includes twenty-ve categories of capital equipment and nine categories of structures.
2581-04_Desai.qxd 1/18/05 13:28 Page 295
show a small and insignicant negative coefcient. This is equally true in
the OLS and the median regression, and the coefcients have similar
magnitudes. In the top panel, the estimate in column 3-1 indicates that an
asset type whose log investment grew by 1 standard deviation (a log value
of 0.36) more than the median asset from 1994 to 1999 would be expected
to have a drop in log investment from 2000 to 2002 about one-sixth of a
standard deviation larger than that of the median rm.
Column 3-2 repeats this analysis but splits the data into the early and
late periods of the boom, 199497 and 199799, respectively. Here,
although the coefcient estimates are imprecise, they are not consistent
with the typical overhang story. If anything, the coefcients are again
larger in absolute value in the earlier period than in the later period. Indeed,
both of the point estimates in the later period are greater than zero, suggesting
that those assets whose real investment grew most in the 1990s saw even
greater investment growth in the 2000s. The irrational exuberance
hypothesis would say just the opposite. Of course, in both cases the
regressions do not control for anything but merely indicate the absence of
a strong negative correlation. Using the rm-level data, we can further
investigate these phenomena at the rm level, with better controls for
observables related to investment opportunities.
Evidence at the Firm Level
Our rm-level sample includes all companies that appear in the Com-
pustat research file from 1962 to 2003. Figure 3 plots the average
investment rate (dened as capital expenditure divided by the beginning-
of-period net capital stock) for manufacturing rms, for nonmanufactur-
ing rms, and for rms involved in information businesses. Information
businesses are defined as those in NAICS categories 334 (computer
manufacturing) and 51 (information); this grouping is one we return to
later, because the irrational exuberance of the late 1990s is commonly
viewed as having been most extreme there. These data reveal the same
pattern as the aggregate data: investment rates rose dramatically in the
1990s and then fell dramatically in the 2000s. We cannot say how repre-
sentative the universe of publicly traded firms is of the rest of the econ-
omy, but in some ways the sheer magnitude of the firm-level sample
makes it an overwhelmingly important component of aggregate invest-
ment on its own. Our calculations suggest that aggregate capital expendi-
296 Brookings Papers on Economic Activity, 2:2004
2581-04_Desai.qxd 1/18/05 13:28 Page 296
ture in the rms in the Compustat data constituted 85 to 90 percent of pri-
vate, nonresidential investment in the United States for most of the last
twenty-ve years.
18
Our sample in 2003 does not include all rms, since
some share of rms had yet to have their reports coded by Compustat at
the time of our analysis. Nonetheless, the sample is large (more than 80
percent of the 2002 sample) and provides a perspective not afforded by
the industry-or asset-level data, given their earlier cutoffs.
We begin with the general evidence that parallels the previous results
in examining the changes in investment rates during the bust and the sub-
sequent boom, but with the advantage that, in the rm-level data, we can
compute the change in the investment rate because we have capital stock
data for each rm. Our modied regression equation, then, is
Mihir A. Desai and Austan D. Goolsbee 297
18. One important shortcoming of the Compustat data (and common to virtually all
empirical work that uses these data to study investment) is the inability to isolate domestic
from international expenditure or the degree to which q measures worldwide rather than
domestic investment opportunities.
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1964 1968 1972 1976 1980 1984 1988 1992 1996 2000
Nonmanufacturing
Manufacturing
Information businesses
Ratio
a
Source: Authors calculations using Compustat data.
a. Ratio of current capital expenditure to the capital stock in the previous period, calculated as described in appendix A.
Figure 3. Firm-Level Investment Rates by Sector, 19622003
2581-04_Desai.qxd 1/18/05 13:28 Page 297
where I is capital expenditure at the rm level and is scaled by the lagged
capital stock K.
19
As before, the top panel of table 4 presents the OLS results and the bot-
tom panel the median regression results. Column 4-1 reports results for a
specication that emphasizes the relationship between the change in
investment rates over the 199499 boom and that over the 200002 bust.
Given that a rm must have existed in 1994, 1999, 2000, and 2002 to
appear in the sample for this regression, the sample size is somewhat
( ) , 2
2002 2000 1999 1994
I
K
I
K
I
K
I
K
i i i i
_
,
_
,
+
_
,
_
,
1
]
1
+ e
298 Brookings Papers on Economic Activity, 2:2004
19. Appendix A describes how we compute the capital stock for each rm, following
Salinger and Summers (1984) and Cummins, Hassett, and Hubbard (1994).
Table 4. Regressions of Changes in Investment Rates during 200002 on Changes in
Investment Rates during the 1990s, Using Data by Firm
a
Independent variable 4-1 4-2 4-3
Ordinary least squares regressions
Investment rate change, 199499 0.0325 0.0237
(0.0234) (0.0216)
Investment rate change, 199799 0.0739
(0.0376)
Investment rate change, 199497 0.0174
(0.0252)
Change in sales, 200002
b
0.0364
(0.0304)
No. of observations 3,249 3,225 3,172
Adjusted R
2
.002 .005 .004
Median regressions
Investment rate change, 199499 0.0170 0.0120
(0.0051) (0.0051)
Investment rate change, 199799 0.0351
(0.0063)
Investment rate change, 199497 0.0071
(0.0051)
Change in sales, 200002
b
0.0545
(0.0040)
No. of observations 3,249 3,225 1,798
Source: Authors regressions using data from Compustat.
a. The dependent variable in all regressions is the change in the rms investment rate, as dened in the text, from 2000 to
2002. Numbers in parentheses are standard errors.
b. In percent.
2581-04_Desai.qxd 1/18/05 13:28 Page 298
smaller than the full universe of rms. These results again show a very
small and insignicant negative correlation in the changes in investment
rates. The median percentage change in the capital stock from 2000 to
2002 was 0.3 percent. The estimated coefcients are indeed tiny com-
pared with the median rm. The coefcient on the lagged investment
change variable indicates that a rm whose increase in investment rate
during the boom was 1 standard deviation (about 0.65) above that of the
median rm would have seen its investment fall by about 0.02, or only
about one-thirty-fth of a standard deviation, during the bust. The bottom
panel of table 4 repeats this specication but controls for outliers by using
a median regression (particularly important when rm data are used), and
here the coefcient is even smaller but statistically significant. A firm
whose investment grew by 1 standard deviation more than the median dur-
ing the boom would have seen its investment fall during the bust by only
about one-seventieth of a standard deviation more than the median rm.
The regressions reported in column 4-2 of both panels in table 4 split
the 199499 period into two parts as an additional control, to account for
rms whose size is trending upward. Inclusion of this split, however, pro-
duced little change in the general result of a very small negative impact
relative to trend. Finally, the regressions in column 4-3 include the per-
centage change in the rms real sales as an additional control, to further
take into account the fact that rms might be growing or shrinking over
the period in a way that drives the investment results. (Recall the large
coefcients on manufacturing investment in the industry-level data.) In
this regression, sales growth is positively correlated with growth in
investment, but the evidence on reversion is even a bit more modest than
in the rst two columns.
The evidence thus far, then, provides very limited support for the view
that rms, asset types, and industries that had major increases in their
investment in the 1990s experienced major drops in investment in the
2000s. This seems to suggest that overhang was not the dominant factor
inuencing investment in the later period. A more precise test is available,
however, by relating overhang to the sensitivity of investment to funda-
mentals at the rm level.
Evidence on Overhang and the Sensitivity of Investment
The rm-level data allow us to further examine whether rms have
been less responsive to changes in tax-adjusted q in the 2000s if they had
Mihir A. Desai and Austan D. Goolsbee 299
2581-04_Desai.qxd 1/18/05 13:28 Page 299
signicant valuation increases in the 1990s. If, in fact, rms experiencing
large changes in market value exhibit a different response than other rms
to tax-adjusted q in the 2000s, this could help explain why taxes have not
seemed to have a major impact on investment. The next section and
appendix B discuss in more detail our tax-adjusted measure of q and the
model underlying it. There we provide a fuller discussion of the measure-
ment issues and the predictions of that model, but we include this analysis
here in order to fully address the overhang phenomenon. Our basic esti-
mating equation will add an interaction term to the standard equation
relating investment to q.
We investigate the relevance of two different measures of overhang
in the 1990s: one based on equity values and one based on capital expan-
sion. Table 5 reports results of regressions using the lagged change in q
as a measure of the degree to which overhang is operative.
20
We create
the variable
t 7
t 3
q
2000+
it
, which is the change in q observed in the period
three to seven years before the current year and only for the
time period 200003.
21
So, in 2002 for example, this variable would be
the change in the firms q from 1995 to 1999. Before the 2000s this
variable is always zero. One view of the overhang hypothesis is that
investment for firms with large capital overhangs from the 1990s should
be less sensitive to fundamentals or tax rates.
22
This yields the following investment equation:
(3) (I/K)
it
i
+
t
+
1
Q
it
+ Q
it
(
t 7
t 3
q
2000+
it
) +
2
(cash/K)
it
+
it
.
Here
i
and
t
are rm and year dummies, respectively, and Q is tax-
adjusted q, as dened below. Column 5-1 of table 5 presents the results
300 Brookings Papers on Economic Activity, 2:2004
20. We considered using the lagged change in the price-earnings ratio as the measure
of rms with overhang, but this had the obvious problem that many rms had negative
earnings.
21. Other lags, such as the change in q from ve years ago to two years ago, yield sim-
ilar results.
22. In a previous draft we also examined whether having had a large increase in K or in
q during the 1990s led the level of investment at the rm to be lower, controlling for current
q (as opposed to the increase changing the slope of the investment-q relationship). We
found virtually no evidence that it did.
2581-04_Desai.qxd 1/18/05 13:28 Page 300
T
a
b
l
e
5
.
R
e
g
r
e
s
s
i
o
n
s
T
e
s
t
i
n
g
S
e
n
s
i
t
i
v
i
t
y
o
f
t
h
e
F
i
r
m
-
L
e
v
e
l
I
n
v
e
s
t
m
e
n
t
-
Q
R
e
l
a
t
i
o
n
s
h
i
p
t
o
P
a
s
t
C
h
a
n
g
e
s
i
n
q
a
I
n
f
o
r
m
a
t
i
o
n
A
l
l
A
l
l
b
u
s
i
n
e
s
s
e
s
M
a
n
u
f
a
c
t
u
r
i
n
g
A
l
l
i
n
d
u
s
t
r
i
e
s
i
n
d
u
s
t
r
i
e
s
o
n
l
y
o
n
l
y
i
n
d
u
s
t
r
i
e
s
I
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
5
-
1
5
-
2
5
-
3
5
-
4
5
-
5
Q
0
.
0
1
2
4
0
.
0
2
2
1
0
.
0
1
4
3
0
.
0
1
3
7
0
.
0
1
2
3
(
0
.
0
0
1
4
)
(
0
.
0
0
1
3
)
(
0
.
0
0
3
2
)
(
0
.
0
0
1
8
)
(
0
.
0
0
1
4
)
Q
q
(
t
3
t
o
t
7
)
,
0
.
0
0
0
6
0
.
0
0
0
2
0
.
0
0
0
5
0
.
0
0
1
0
0
.
0
0
0
6
2
0
0
0
o
r
a
f
t
e
r
b
(
0
.
0
0
0
4
)
(
0
.
0
0
0
5
)
(
0
.
0
0
1
1
)
(
0
.
0
0
0
7
)
(
0
.
0
0
0
4
)
Q
q
(
t
3
t
o
t
7
)
,
0
.
0
0
0
3
1
9
9
9
o
r
b
e
f
o
r
e
c
(
0
.
0
0
0
3
)
R
a
t
i
o
o
f
c
a
s
h
o
w
t
o
0
.
0
1
9
8
0
.
0
1
5
7
0
.
0
0
6
9
0
.
0
2
3
6
0
.
0
1
9
8
c
a
p
i
t
a
l
s
t
o
c
k
(
0
.
0
0
3
1
)
(
0
.
0
0
2
4
)
(
0
.
0
0
5
7
)
(
0
.
0
0
4
2
)
(
0
.
0
0
3
1
)
Y
e
a
r
d
u
m
m
i
e
s
i
n
c
l
u
d
e
d
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
F
i
r
m
d
u
m
m
i
e
s
i
n
c
l
u
d
e
d
Y
e
s
N
o
Y
e
s
Y
e
s
Y
e
s
N
o
.
o
f
o
b
s
e
r
v
a
t
i
o
n
s
6
9
,
5
4
0
6
9
,
5
4
0
1
1
,
7
5
8
3
6
,
3
1
3
6
9
,
5
4
0
A
d
j
u
s
t
e
d
R
2
.
4
0
3
.
0
3
9
.
3
9
9
.
3
7
7
.
4
0
3
S
o
u
r
c
e
:
A
u
t
h
o
r
s
r
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
d
a
t
a
f
r
o
m
C
o
m
p
u
s
t
a
t
.
a
.
T
h
e
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
n
a
l
l
r
e
g
r
e
s
s
i
o
n
s
i
s
t
h
e
r
m
s
i
n
v
e
s
t
m
e
n
t
r
a
t
e
i
n
a
g
i
v
e
n
y
e
a
r
,
a
s
d
e
n
e
d
i
n
t
h
e
t
e
x
t
.
Q
i
s
t
a
x
-
a
d
j
u
s
t
e
d
q
,
d
e
n
e
d
a
s
[
q
/
(
1
)
]
[
(
1
)
/
(
1
)
]
,
w
h
e
r
e
i
s
t
h
e
s
t
a
n
d
a
r
d
m
e
a
-
s
u
r
e
o
f
t
h
e
t
a
x
t
r
e
a
t
m
e
n
t
o
f
i
n
v
e
s
t
m
e
n
t
a
n
d
i
s
t
h
e
c
o
r
p
o
r
a
t
e
t
a
x
r
a
t
e
.
N
u
m
b
e
r
s
i
n
p
a
r
e
n
t
h
e
s
e
s
a
r
e
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
a
t
t
h
e
r
m
l
e
v
e
l
.
b
.
Q
t
i
m
e
s
t
h
e
c
h
a
n
g
e
i
n
q
o
b
s
e
r
v
e
d
o
v
e
r
t
h
e
p
e
r
i
o
d
t
h
r
e
e
t
o
s
e
v
e
n
y
e
a
r
s
b
e
f
o
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
,
w
h
e
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
i
s
2
0
0
0
,
2
0
0
1
,
2
0
0
2
,
o
r
2
0
0
3
.
c
.
A
s
a
b
o
v
e
,
w
h
e
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
i
s
1
9
9
9
o
r
b
e
f
o
r
e
.
2581-04_Desai.qxd 1/18/05 13:28 Page 301
from estimating this equation over all rms. It shows that there is no sig-
nicant difference in the investment-q relationship in the 2000s for rms
that had larger run-ups in their stock prices in the 1990s. Indeed, the point
estimate is actually positive, although small. Column 5-2 excludes the
rm dummies, so that we are explicitly comparing results across rms
rather than within a given rm. The result on the interaction term is very
similar to that in column 5-1: positive and not signicant.
Column 5-3 returns to the specication with rm dummies but restricts
the sample to rms in information businesses; these are the rms most
closely associated with the technology bubble. There is again no evidence
that large increases in equity values in the 1990s have reduced the sensi-
tivity of investment to the fundamentals in the 2000s. The point estimate
on the interaction term is insignicant, although this time slightly less
than zero. Column 5-4 repeats the analysis, this time for manufacturing
rms only, and again there is nothing notable. Finally, column 5-5 inves-
tigates whether the relationship changed any differently in the 2000s than
it did in earlier periods that followed asset price increases. The evidence
suggests that it did not.
Table 6 repeats the exercise reported in table 5 but uses the lagged per-
centage change in capital for the rm during the 1990s as the measure of
overhang. The advantage of the lagged change in q as the overhang mea-
sure in table 5 is that it picks up more directly the inuence of asset price
bubbles, which typically underlie the popular explanation of overhang.
The lagged percentage change in the capital stock as used in table 6, in
contrast, is a more direct measure of capital accumulation.
Table 6 reports estimates of the following equation:
(4) (I/K)
it
i
+
t
+
1
Q
it
+ Q
it
(%
t 7
t 3
K
2000+
it
) +
2
(cash/K)
it
+
it
,
where %
t 7
t 3
K
2000+
it
is the percentage change in the net capital stock of
the firm between time t 3 and time t 7 for 200003 (in other words,
the change in the capital stock during the mid-1990s).
Estimating this equation for the entire sample of rms, as reported in
column 6-1, does show a signicant negative coefcient on the interacted
Q term, indicating that rms that had larger accumulations of capital in
the 1990s did, indeed, show less sensitivity to the fundamentals in their
investment behavior in the 2000s. Although the direction is consistent
with the overhang view, the magnitude is extremely small. To see this,
consider that the highest mean value of lagged capital growth was 1.37 in
302 Brookings Papers on Economic Activity, 2:2004
2581-04_Desai.qxd 1/18/05 13:28 Page 302
T
a
b
l
e
6
.
R
e
g
r
e
s
s
i
o
n
s
T
e
s
t
i
n
g
S
e
n
s
i
t
i
v
i
t
y
o
f
t
h
e
F
i
r
m
-
L
e
v
e
l
I
n
v
e
s
t
m
e
n
t
-
Q
R
e
l
a
t
i
o
n
s
h
i
p
t
o
P
a
s
t
C
h
a
n
g
e
s
i
n
I
n
v
e
s
t
m
e
n
t
a
I
n
f
o
r
m
a
t
i
o
n
A
l
l
A
l
l
b
u
s
i
n
e
s
s
e
s
M
a
n
u
f
a
c
t
u
r
i
n
g
A
l
l
i
n
d
u
s
t
r
i
e
s
i
n
d
u
s
t
r
i
e
s
o
n
l
y
o
n
l
y
i
n
d
u
s
t
r
i
e
s
I
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
6
-
1
6
-
2
6
-
3
6
-
4
6
-
5
q
0
.
0
1
2
4
0
.
0
2
1
4
0
.
0
1
5
5
0
.
0
1
1
7
0
.
0
1
3
7
(
0
.
0
0
1
4
)
(
0
.
0
0
1
2
)
(
0
.
0
0
1
1
)
(
0
.
0
0
1
7
)
(
0
.
0
0
1
5
)
Q
K
(
t
3
t
o
t
7
)
,
0
.
0
0
2
7
0
.
0
0
0
8
0
.
0
0
2
1
0
.
0
0
2
3
0
.
0
0
3
0
2
0
0
0
o
r
a
f
t
e
r
b
(
0
.
0
0
0
7
)
(
0
.
0
0
0
4
)
(
0
.
0
0
1
1
)
(
0
.
0
0
0
9
)
(
0
.
0
0
0
7
)
Q
K
(
t
3
t
o
t
7
)
,
0
.
0
0
1
3
1
9
9
9
o
r
b
e
f
o
r
e
c
(
0
.
0
0
0
4
)
R
a
t
i
o
o
f
c
a
s
h
o
w
t
o
0
.
0
1
6
8
0
.
0
1
1
4
0
.
0
0
5
8
0
.
0
1
4
3
0
.
0
1
6
8
c
a
p
i
t
a
l
s
t
o
c
k
(
0
.
0
0
2
4
)
(
0
.
0
0
1
7
)
(
0
.
0
0
4
5
)
(
0
.
0
0
3
3
)
(
0
.
0
0
2
4
)
Y
e
a
r
d
u
m
m
i
e
s
i
n
c
l
u
d
e
d
Y
e
s
Y
e
s
Y
e
s
Y
e
s
Y
e
s
F
i
r
m
d
u
m
m
i
e
s
i
n
c
l
u
d
e
d
Y
e
s
N
o
Y
e
s
Y
e
s
Y
e
s
N
o
.
o
f
o
b
s
e
r
v
a
t
i
o
n
s
8
3
,
1
4
7
8
3
,
1
4
7
1
4
,
7
3
5
4
4
,
3
2
6
8
3
,
1
4
7
A
d
j
u
s
t
e
d
R
2
.
3
7
1
.
0
3
5
.
3
5
7
.
3
3
7
.
3
7
1
S
o
u
r
c
e
:
A
u
t
h
o
r
s
r
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
d
a
t
a
f
r
o
m
C
o
m
p
u
s
t
a
t
.
a
.
T
h
e
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
n
a
l
l
r
e
g
r
e
s
s
i
o
n
s
i
s
t
h
e
r
m
s
i
n
v
e
s
t
m
e
n
t
r
a
t
e
,
a
s
d
e
n
e
d
i
n
t
h
e
t
e
x
t
,
i
n
a
g
i
v
e
n
y
e
a
r
.
Q
i
s
t
a
x
-
a
d
j
u
s
t
e
d
q
a
s
d
e
n
e
d
i
n
t
a
b
l
e
5
;
K
i
s
t
h
e
c
a
p
i
t
a
l
s
t
o
c
k
.
N
u
m
b
e
r
s
i
n
p
a
r
e
n
t
h
e
s
e
s
a
r
e
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
a
t
t
h
e
r
m
l
e
v
e
l
.
b
.
Q
t
i
m
e
s
t
h
e
c
h
a
n
g
e
i
n
t
h
e
c
a
p
i
t
a
l
s
t
o
c
k
o
b
s
e
r
v
e
d
o
v
e
r
t
h
e
p
e
r
i
o
d
t
h
r
e
e
t
o
s
e
v
e
n
y
e
a
r
s
b
e
f
o
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
,
w
h
e
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
i
s
2
0
0
0
,
2
0
0
1
,
2
0
0
2
,
o
r
2
0
0
3
.
c
.
A
s
a
b
o
v
e
,
w
h
e
r
e
t
h
e
c
u
r
r
e
n
t
y
e
a
r
i
s
1
9
9
9
o
r
b
e
f
o
r
e
.
2581-04_Desai.qxd 1/18/05 13:28 Page 303
2002 (with a median value of past growth of 0.41). This value predicts
that the coefcient on Q would fall by only 0.0037 (and only 0.0011 for
the median). When we explicitly compare across rms by dropping the
rm dummies (column 6-2), the point estimate becomes positive. Column
6-3 restricts the sample to information businesses as before; the coef-
cient on lagged capital growth, although slightly negative, is similarly
modest. Column 6-4 repeats the analysis for manufacturing rms only
and again nds similar results. Column 6-5 demonstrates that, with this
measure of overhang, there is normally a small negative impact of lagged
capital growth on current investment rates, even in the period before the
2000s. The difference in the coefcient between the 2000s and the pre-
2000s period is only about 0.0017.
Taken together, the results in this section provide little evidence that
capital overhang has played a key role in investment behavior in the
2000s. Low investment during the bust is not correlated strongly with
excessive investment in the boom. Similarly, the sensitivity of invest-
ment in the 2000s to the fundamentals is not markedly different for
firms overall or for those firms usually at the heart of the overhang
view. In other words, the standard firm-level model using tax-adjusted q
has not become noticeably worse at explaining investment. Accord-
ingly, in the following section we use this model to analyze the impact
of taxes.
q Theory, Investment Incentives, and Dividend Taxes:
Theory and Empirics
As a prelude to using the tax-adjusted q model to study the impact of
the Bush tax cuts, it is useful to consider the aggregate movements in q
over the whole of the period that our rm sample covers, 19622003, in
thinking about the root determinants of the behavior of aggregate invest-
ment in the 1990s and 2000s. Figure 4 plots average q for the corporate
sector as a whole, measured as the total market value of all publicly traded
rms as computed by CRSP (Center for Research in Security Prices)
divided by the total stock of corporate capital as computed by the BEA in
its xed reproducible tangible wealth series. The series shows a historic
rise in q in the mid-1990s and an unprecedentedly steep fall in the
304 Brookings Papers on Economic Activity, 2:2004
2581-04_Desai.qxd 1/18/05 13:28 Page 304
Mihir A. Desai and Austan D. Goolsbee 305
23. A plot using the average of rm qs in our sample (not shown) yields a similar
picture.
2000s.
23
Previous studies have found only very small coefcients on tax-
adjusted q, and so this rise and fall in q might not imply much in the way
of aggregate investment changes. We estimate the investment relation-
ship, however, within a slightly different framework than is typical, to
overcome possibly conating measurement issues. It should be clear that,
if the true coefcient on q for investment is not 0.02 but closer to 1, as
argued below, the investment collapse is eminently comprehensible
within the conventional framework. Clearly, within a standard q model of
investment, an equity price bubble can still drive investment up and then
down through movements in q. Such an account of the investment experi-
ence of the 1990s and 2000s is distinct from the intuition of a lingering
overhang from the 1990s.
Of course, as critics have frequently pointed out, the fundamental
question is why the coefcients on q in investment regressions are typi-
cally so low, implying extremely large adjustment costs. One of the key
0.5
1.0
1.5
2.0
1964 1968 1972 1976 1980 1984 1988 1992 1996 2000
Ratio
a
Source: Authors calculations using BEA and Compustat data.
a. Ratio of the aggregate market value of all rms in the Compustat sample to the stock of corporate capital as computed by
the BEA in its Fixed Reproducible Tangible Wealth series.
Figure 4. Aggregate q, 19622002
2581-04_Desai.qxd 1/18/05 13:28 Page 305
potential problems discussed in the literature has been the importance of
measurement error in q. Marginal q is the variable of interest, but the data
can provide only average q. At least some of the existing literature has
argued that measurement error is at the root of the relatively weak empir-
ical performance of traditional investment models.
24
This issue of mea-
surement error is particularly important for thinking about the impact of
taxation, as we demonstrate below.
If, in fact, an empirical implementation of the q model provides more
reasonable coefcients through an alternative strategy of dealing with
these measurement issues, then these estimates can serve as the founda-
tion for analyses of the true marginal costs of adjustment and the impact
of the various tax policy changes enacted during the Bush administration
(the changes in depreciation allowances as well as the changes in divi-
dend and capital gains taxes). The rest of this section undertakes such an
analysis.
Tax-Adjusted q Theory, Dividend Taxes, and the Marginal Source
of Funds
To use the q model to analyze the impact of taxes, particularly divi-
dend taxes, we revisit the techniques for incorporating taxes into such
models developed by Summers and by James Poterba.
25
A crucial issue in
determining the impact of dividend taxes in this framework is what the
marginal source of funds for rms investments is. Briey, if the marginal
source of funds is retained earnings, then dividend taxes will have no
impact on marginal investment incentives. But if the marginal source of
funds is new equity, then dividend taxes will inuence investment. This
distinction is the subject of an enduring debate in public nance between
the new and the traditional view of dividend taxation.
Appendix B works through the implications of the two views in some
detail and provides alternative estimating equations. The investment
model typically estimated in the literature follows the traditional view. In
this view dividend taxes influence investment by, essentially, double-
taxing corporate income. Assuming quadratic adjustment costs, this view
generates the following investment-q relationship:
306 Brookings Papers on Economic Activity, 2:2004
24. See, for example, Cummins, Hassett, and Hubbard (1994), Bond and Cummins
(2000), and Goolsbee (2001).
25. Summers (1981); Poterba and Summers (1983, 1985).
2581-04_Desai.qxd 1/18/05 13:28 Page 306
where I is investment, K is the rms capital stock, is the adjustment
cost parameter, and represents an average investment rate. Under this
assumption, net new equity nances investment, so that investment is
determined by the point at which shareholders are indifferent between
holding a dollar inside or holding it outside the rm. In a world without
other taxes, the rm stops investing once q 1. Of course, if investment is
heavily subsidized (that is, if > ), rms may even continue investing
with q < 1, but the general idea is the same. Changes in dividend taxes
will inuence equity values and investment incentives by inuencing the
relative preference of investors to hold their money inside rather than out-
side the rm.
If, however, retained earnings are the marginal source of nance, the
traditional investment-q relationship of equation 5 will not hold. In this
case (again assuming quadratic adjustment costs), the relationship will
follow
where is the tax rate on dividends is and c is the accrual-equivalent tax
rate on capital gains.
Equation 6 corresponds to the new (or trapped equity or tax capi-
talization) view of the role of dividend taxation. In this view dividend
taxes do not inuence the tax term for marginal investments.
26
Instead
they are fully capitalized into existing share prices. In other words,
changes in dividend taxes serve solely as a penalty or windfall on existing
rm values. To see the intuition behind this, consider a rm that uses
retained earnings at the margin to nance investment, with dividends
determined as a residual. In this model dividends are the only means of
distributing earnings to shareholders. In this setting, given that retained
( ) , 6
1
1
1
1
1
1
1
I
K
q
c
t
t
t
_
,
_
,
1
]
1
1
1
_
,
( ) , 5
1
1
1
1
1
I
K
q
t
t
t
_
,
_
,
c
308 Brookings Papers on Economic Activity, 2:2004
27. See the work of Poterba and Summers (1985), Chetty and Saez (2004), and Poterba
(2004) and the papers they cite. Auerbach (2002) points out, however, that interpretation of
the empirical evidence on this point is complicated by the fact that temporary cuts in divi-
dend taxes should encourage dividend payouts, even under the new view.
28. Fuller assessments of both views can be found in Auerbach and Hassett (2003),
Carroll, Hassett, and Mackie (2003), and Poterba and Summers (1985).
2581-04_Desai.qxd 1/18/05 13:28 Page 308
dom by Poterba and Summers,
29
there have been no direct attempts to test
between the two views using investment data.
Poterba and Summers use thirty years of annual data from the United
Kingdom to test between equations 5 and 6; their results support the tradi-
tional view. Although Alan Auerbach and Hassett have been critical of
these ndings for, among other things, failing to account for other macro-
economic and tax changes occurring at the same time,
30
these estimates
are still the only direct empirical tests of how dividend taxes affect invest-
ment. Oddly, no one has extended the methods of Poterba and Summers
to firm-level data, where it is possible to control for many aggregate
factors. Nor has anyone ever applied their method to the United States,
perhaps because, until the 2003 tax cut, U.S. dividend taxes did not
change in isolation from other changes, but rather varied only through
changes in personal income tax rates. Instead the empirical work testing
the new versus the traditional view has adopted the indirect method of
examining the relationship between dividend taxes and dividend pay-
ments (or the valuation of dividend payments by investors).
31
The recent large changes in dividend taxes, however, make such an
analysis possible. Indeed, testing between these two views (and making
the required detour into the public nance debate) is critical for evaluating
the impact of the Bush tax cuts on investment. If the marginal source of
funds turns out to be retained earnings, the dividend tax cut will have little
or no impact on marginal incentives to invest. Before explicitly testing
between the two views, however, we lay out the basic tax-adjusted q
model and illustrate why we believe measurement error is a primary reason
that such models have implied high adjustment costs and have performed
so poorly in the past.
Empirical Implementation of the Q Model
In computing q empirically, we use the historical and current Compu-
stat database, which provides data on a panel of rms from 1962 through
Mihir A. Desai and Austan D. Goolsbee 309
29. Poterba and Summers (1983, 1985).
30. Auerbach (2002); Auerbach and Hassett (2003).
31. See, for example, Bernheim and Wantz (1995), Poterba (2004), Chetty and Saez
(2004), and Poterba and Summers (1985), as well as the opposing evidence in Bolster and
Janjigian (1991), Blouin, Raedy, and Shackelford (2004), and Ikenberry and Julio (2004).
2581-04_Desai.qxd 1/18/05 13:28 Page 309
2003.
32
For some rms we also match this sample to the earnings esti-
mates provided by I/B/E/S (Institutional Brokers Estimate System). Esti-
mates of the tax term at the asset level are derived from data provided
generously by Dale Jorgenson.
33
As described in more detail in appendix A,
we follow Cummins, Hassett, and Hubbard and Robert Chirinko, Steven
Fazzari, and Andrew Meyer in using the BEAs capital flows table for
1997 to calculate the share of investment in each industry for each asset
type.
34
With that weighting, we calculate the weighted-average tax term in
each year for each four-digit industry in the Compustat data. Average
marginal tax rates on dividend and capital gains income (on an accrual
basis) are taken from Poterba.
35
Further discussion of the variable con-
struction and the sources of data is provided in appendix A.
The measurement of q and, in turn, Q, hinges on constructing a measure
of the ratio of a rms market value to its book value. The corporate nance
literature and the public nance literature have diverged somewhat in their
measurement of this ratio, and we consider both alternatives in the results
that follow. The corporate nance literature, as exemplied in the 1997
paper by Steven Kaplan and Luigi Zingales,
36
employs data from Compu-
stat to derive a measure of q as ,
where BV stands for book value and MV for market value, and all values
are taken from public financial records.
37
In contrast, the public finance
literature has emphasized the derivation by Salinger and Summers,
38
which constructs q as , where debt and equity
values are taken from nancial reports, but the market value of assets is
imputed using perpetual inventory methods and valuations of inventory as
discussed in the data appendix to Cummins, Hassett, and Hubbard.
39
MV Equity MV Debt
MV Assets
+
BV Assets MV Equity BV Equity
BV Assets
+
( )
310 Brookings Papers on Economic Activity, 2:2004
32. Because of reporting conventions, the 2003 sample is somewhat smaller than the
sample in earlier years.
33. The data are described in more detail in Jorgenson and Yun (2001). Importantly,
the Jorgenson calculations do not take any future expectations of tax changes into account.
They use only the statutory tax rules for the year in question.
34. Cummins, Hassett, and Hubbard (1994); Chirinko, Fazzari, and Meyer (1999).
35. Poterba (2004).
36. Kaplan and Zingales (1997).
37. This numerator is also sometimes adjusted for deferred taxes.
38. Salinger and Summers (1984); see also Cummins, Hassett, and Hubbard (1994).
39. Cummins, Hassett, and Hubbard (1994).
2581-04_Desai.qxd 1/18/05 13:28 Page 310
Implicitly, this formulation takes the market value of debt to be its book
value.
As with any rm-level analysis employing Compustat data to study
investment, rules for considering extreme observations must be employed.
Following studies such as that by Simon Gilchrist and Charles Himmel-
berg, we truncate our measures of q (and of investment and cash ow as a
share of the capital stock) at the 1st and the 99th percentile.
40
Investment
rates and cash ow rates are taken as the ratio of capital expenditure and
operating cash ow before depreciation, respectively, to the capital stock.
Table 7 reports the results of estimating q models in our rm sample,
under both of the two alternative denitions of q, as well as Q, that is,
q adjusted for taxes, . We postpone discussion of the relevance
of dividend taxes, and so our estimating equation is equation 5 above,
with and without consideration of taxes. Columns 7-1 and 7-4 of table 7
contrast the performance of the corporate finance and public finance
measures of q without consideration of tax factors. Both coefcients are
signicant and positive, but the coefcient on the corporate nance q is
much larger. Inspection of the public nance qs indicates that extreme
values make up a large fraction of the sample and may contribute to this
pattern. Comparison of columns 7-2 and 7-5 provides a similar result,
with significantly larger coefficients on the corporate financebased
measure of Q. Nonetheless, the coefcients reect the common difculty
in this literature, which is that these small coefcients translate into
extremely high adjustment cost parameters (the inverse of the measured
coefcient). Inclusion of both q and Q, in the specications reported in
columns 7-3 and 7-6, results in a similar pattern, but does indicate that tax-
adjusted q outperforms q in explaining investment. This nding parallels
the nding by Summers of the relevance of tax adjustments in improving
the estimation of the q model.
41
Given the relative performance of q and Q in the results reported in
table 7, it is useful to consider separately the terms that make up Q to
better understand the sources of the relatively small coefcients on Q. As
discussed above, , and so the specications in table 7 can
Q
q
t
1
1
1
q
t
1
1
1
_
,
n
i
t
i
o
n
s
a
R
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
c
o
r
p
o
r
a
t
e
n
a
n
c
e
R
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
p
u
b
l
i
c
n
a
n
c
e
d
e
n
i
t
i
o
n
o
f
q
d
e
n
i
t
i
o
n
o
f
q
I
n
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
7
-
1
7
-
2
7
-
3
7
-
4
7
-
5
7
-
6
q
0
.
0
3
7
9
0
.
1
1
1
1
0
.
0
0
0
7
0
.
0
0
3
0
(
0
.
0
0
1
9
)
(
0
.
0
1
7
4
)
(
0
.
0
0
0
2
)
(
0
.
0
0
1
8
)
Q
0
.
0
2
3
1
0
.
0
8
6
3
0
.
0
0
0
5
0
.
0
0
2
3
(
0
.
0
0
1
1
)
(
0
.
0
1
0
2
)
(
0
.
0
0
0
1
)
(
0
.
0
0
1
0
)
R
a
t
i
o
o
f
c
a
s
h
o
w
t
o
0
.
0
0
2
0
0
.
0
0
0
6
0
.
0
0
1
5
0
.
0
0
0
3
0
.
0
0
1
3
0
.
0
0
1
3
c
a
p
i
t
a
l
s
t
o
c
k
(
0
.
0
0
1
5
)
(
0
.
0
0
1
5
)
(
0
.
0
0
1
6
)
(
0
.
0
0
1
5
)
(
0
.
0
0
1
5
)
(
0
.
0
0
1
6
)
N
o
.
o
f
o
b
s
e
r
v
a
t
i
o
n
s
1
6
0
,
0
5
1
1
4
2
,
0
4
3
1
4
2
,
0
4
3
1
6
1
,
4
1
6
1
4
2
,
8
8
2
1
4
2
,
8
8
2
A
d
j
u
s
t
e
d
R
2
.
3
7
7
.
3
7
6
.
3
7
7
.
3
6
8
.
3
6
7
.
3
6
7
S
o
u
r
c
e
:
A
u
t
h
o
r
s
r
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
d
a
t
a
f
r
o
m
C
o
m
p
u
s
t
a
t
.
a
.
T
h
e
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
n
a
l
l
r
e
g
r
e
s
s
i
o
n
s
i
s
t
h
e
r
m
s
i
n
v
e
s
t
m
e
n
t
r
a
t
e
,
a
s
d
e
n
e
d
i
n
t
h
e
t
e
x
t
,
i
n
a
g
i
v
e
n
y
e
a
r
.
Q
i
s
t
a
x
-
a
d
j
u
s
t
e
d
q
a
s
d
e
n
e
d
i
n
t
a
b
l
e
5
.
A
l
l
r
e
g
r
e
s
s
i
o
n
s
i
n
c
l
u
d
e
y
e
a
r
a
n
d
r
m
d
u
m
m
i
e
s
.
N
u
m
b
e
r
s
i
n
p
a
r
e
n
t
h
e
s
e
s
a
r
e
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
a
t
t
h
e
r
m
l
e
v
e
l
.
2581-04_Desai.qxd 1/18/05 13:28 Page 312
naturally be recast to consider the separate effects for these two terms.
Splitting Q in this manner has the advantage of allowing us to consider
the role of measurement error in biasing the estimates previously
obtained. More specically, Cummins, Hassett, and Hubbard argue that
mismeasurement of q means that using the estimated coefcients from
standard investment regressions can dramatically understate the impact of
investment taxes.
42
They emphasize large tax reforms as being times
when the tax part of Q is not mismeasured, and they use these periods as
the basis for comparing actual with projected investment. The specica-
tions provided in table 7 take a simpler approach but in the same spirit. If
measurement error in q is a problem, splitting Q into two parts has the
advantage that the coefcient (or, more accurately, its absolute value) on
the term should provide a better estimate of the true coefcient on q.
43
Table 8 presents the results from splitting Q into its component parts.
Specically, the specication in the rst column replaces Q with q scaled
by 1 minus the corporate tax rate and terms for the equipment tax term
and the structures tax term. It is difcult to measure a rms relative
investment in equipment and structures, and so we simply include both
tax terms as separate regressors. Given the traditional difculties in
understanding the dynamics of incentives for investment in structures,
44
and given that equipment accounts for approximately 80 percent of cor-
porate investment, we expect the equipment tax term to be much more
precisely estimated. Controls for internal cash ow are included as well.
The key result from this table is that, although the q term remains
small, the coefficient on the equipment tax term is considerably larger
than typically estimated when just using Q and is close to 1 in absolute
value.
45
The second column includes q without a tax adjustment and indi-
cates, as with the results in table 7, that a tax-adjusted q term performs
1
1
t
)
0
.
0
2
3
1
0
.
0
8
5
8
0
.
0
1
6
6
0
.
0
2
4
5
0
.
2
1
2
0
(
0
.
0
0
1
1
)
(
0
.
0
1
0
2
)
(
0
.
0
0
1
1
)
(
0
.
0
0
1
1
)
(
0
.
0
0
5
4
)
q
0
.
1
1
0
3
(
0
.
0
1
7
4
)
T
a
x
t
e
r
m
,
e
q
u
i
p
m
e
n
t
0
.
8
8
9
5
0
.
7
8
6
5
0
.
7
0
7
8
0
.
8
9
4
9
2
.
5
3
5
1
(
0
.
3
1
7
3
)
(
0
.
3
1
6
2
)
(
0
.
2
8
7
0
)
(
0
.
3
1
6
3
)
(
1
.
0
1
0
1
)
T
a
x
t
e
r
m
,
s
t
r
u
c
t
u
r
e
s
0
.
0
1
6
9
0
.
0
0
6
4
0
.
0
3
3
3
0
.
0
1
2
7
0
.
7
9
0
2
(
0
.
0
4
5
2
)
(
0
.
0
4
5
3
)
(
0
.
0
4
0
8
)
(
0
.
0
4
5
0
)
(
0
.
2
7
1
6
)
R
a
t
i
o
o
f
c
a
s
h
o
w
t
o
c
a
p
i
t
a
l
s
t
o
c
k
0
.
0
0
0
5
0
.
0
0
0
4
0
.
0
0
9
0
0
.
0
0
0
2
0
.
0
1
7
0
(
0
.
0
0
1
5
)
(
0
.
0
0
1
5
)
(
0
.
0
0
1
8
)
(
0
.
0
0
1
5
)
(
0
.
0
0
2
2
)
N
o
.
o
f
o
b
s
e
r
v
a
t
i
o
n
s
1
4
1
,
6
2
9
1
4
1
,
6
2
9
1
1
1
,
0
5
9
1
4
1
,
2
4
5
4
6
,
1
5
4
A
d
j
u
s
t
e
d
R
2
0
.
3
7
6
.
3
7
7
.
3
8
1
.
3
7
8
S
o
u
r
c
e
:
A
u
t
h
o
r
s
r
e
g
r
e
s
s
i
o
n
s
u
s
i
n
g
d
a
t
a
f
r
o
m
C
o
m
p
u
s
t
a
t
.
a
.
T
h
e
d
e
p
e
n
d
e
n
t
v
a
r
i
a
b
l
e
i
n
a
l
l
r
e
g
r
e
s
s
i
o
n
s
i
s
t
h
e
r
m
s
i
n
v
e
s
t
m
e
n
t
r
a
t
e
,
a
s
d
e
n
e
d
i
n
t
h
e
t
e
x
t
,
i
n
a
g
i
v
e
n
y
e
a
r
.
A
l
l
r
e
g
r
e
s
s
i
o
n
s
i
n
c
l
u
d
e
y
e
a
r
a
n
d
r
m
d
u
m
m
i
e
s
.
N
u
m
b
e
r
s
i
n
p
a
r
e
n
t
h
e
s
e
s
a
r
e
s
t
a
n
d
a
r
d
e
r
r
o
r
s
c
l
u
s
t
e
r
e
d
a
t
t
h
e
r
m
l
e
v
e
l
.
b
.
C
o
n
t
r
o
l
s
f
o
r
t
h
e
p
r
e
s
e
n
t
v
a
l
u
e
o
f
t
a
x
d
e
p
r
e
c
i
a
t
i
o
n
a
l
l
o
w
a
n
c
e
s
o
n
p
r
e
v
i
o
u
s
l
y
p
u
r
c
h
a
s
e
d
i
n
v
e
s
t
m
e
n
t
,
a
s
d
e
s
c
r
i
b
e
d
i
n
a
p
p
e
n
d
i
x
B
.
c
.
C
o
n
t
r
o
l
s
f
o
r
t
h
e
s
h
a
r
e
o
f
d
e
b
t
i
n
r
m
n
a
n
c
i
n
g
.
d
.
F
i
n
a
n
c
i
a
l
a
n
a
l
y
s
t
s
r
e
p
o
r
t
e
d
e
s
t
i
m
a
t
e
s
o
f
t
h
e
r
m
s
e
a
r
n
i
n
g
s
a
r
e
u
s
e
d
a
s
a
n
i
n
s
t
r
u
m
e
n
t
f
o
r
r
m
q
.
2581-04_Desai.qxd 1/18/05 13:28 Page 314
better than ordinary q. In this specification the coefficient on the equip-
ment tax term remains significant and large. The third and fourth
columns report two alternative robustness checks for these results that
are modifications to the basic tax-adjusted q model. First, the theory does
imply that the present value of tax depreciation allowances on previously
purchased investment should be included in the value of the firm. This is
frequently left out of empirical work on Q, since it is difficult to com-
pute. In the third column we approximate the size of these tax shields as
described in appendix A, and we add the value of these shields to the
value of the firm in Q. This does not change the estimated results dra-
matically. We found this to be true for all of the major results in the
paper, and, since computing the allowances requires reducing the sample
by more than 30,000, we exclude them from the results that follow. Sim-
ilarly, the model presented above follows most of the literature in assum-
ing away any issues regarding debt financing. In the fourth column we
incorporate the share of the firms financing that comes from debt,
46
and
the results are again similar.
The large coefcients on the tax term terms are worth dwelling on.
First, the model predicts that these coefcients should be of the same
magnitude as those on tax-adjusted q, but of opposite sign. Here, instead,
the coefcients on the equipment tax term are considerably larger. With
measurement error in q, the coefcient on the tax term may provide a
more realistic estimate of the true coefcient. Such a coefcient is consid-
erably closer to 1 and, consequently, corresponds to more realistic esti-
mates of adjustment costs. Restricting attention to years with major tax
reforms yielded similar estimates.
47
To obtain some further evidence on the role of q mismeasurement as
the reason for the small coefcient on the q term, we also modify the
empirical strategy of Stephen Bond and Cummins,
48
within the frame-
work of table 8. Their intuition is that earnings estimates by equity ana-
lysts as provided in the I/B/E/S database are a part of q that is based only
on fundamentals.
49
Rather than use these estimates to create an alternative
Mihir A. Desai and Austan D. Goolsbee 315
46. Following Summers (1981).
47. These results are similar to ndings by Cummins, Hassett, and Hubbard (1994).
48. Bond and Cummins (2000).
49. Cummins, Hassett, and Oliner (forthcoming) also look at investment equations that
include analysts earnings estimates.
2581-04_Desai.qxd 1/18/05 13:28 Page 315
q measure, however, we use them as instruments for q.
50
Employing the
earnings estimates comes at considerable cost, given the shorter time
frame covered by the I/B/E/S database (19832003) and the severely
restricted number of rms covered by I/B/E/S. Nonetheless, we report in
the fth column of table 8 the results of this estimation. Several points are
worth noting. First, as indicated by the tenfold increase in the coefcient
on tax-adjusted q, mismeasurement of q seems important. Second, the
coefcient on the equipment tax term rises considerably as well. Given
the considerably smaller panel for these instrumental variables results, we
rely on the coefcients on the equipment tax term term in the rst column
of table 8 as the best estimate of the true coefcient from a tax-adjusted
q model. This analysis suggests that the true adjustment costs for invest-
ment are of plausible size, and so we use the model to estimate the impact
of the Bush tax cuts. Finally, it is useful to consider whether the relevance
of the q model is different in manufacturing industries, since many previ-
ous studies have restricted their sample to manufacturing. We prefer not
to do this, since manufacturing accounts for only a small fraction of total
investment. Table 9 replicates the analysis from the rst column of table 8
and divides the sample. Although the reduced sample sizes reduce the
power of these tests, the coefcients on the relevant tax term terms are
quite similar in the two subsamples, suggesting that the model performs
similarly well in both settings.
The Impact of Tax Cuts in the 2000s
During the George W. Bush administration two major changes have
been made to the tax code to reduce taxes on capital. First, in 2003 the top
capital gains tax rate was reduced from 20 percent to 15 percent, and the
tax rate on most dividends was reduced from the ordinary personal
income tax rate (which then had a maximum of 38.6 percent) to the capi-
tal gains tax rate. The second change substantially accelerated deprecia-
tion. In 2002 depreciation allowances for virtually all types of equipment
investment were increased, as rms gained the right to immediately
expense 30 percent of their purchases. In 2003 depreciation allowances
316 Brookings Papers on Economic Activity, 2:2004
50. To be precise, we use the earnings estimates divided by (1 ) as an instrument for
q/(1 ).
2581-04_Desai.qxd 1/18/05 13:28 Page 316
were increased again, as the fraction that could be immediately expensed
increased to 50 percent. Each of these tax changes needs to be treated dif-
ferently under the Q model.
Dividend Taxes
Although implementation of the dividend tax reduction was somewhat
complex, essentially the maximum rate on dividends for individuals was
reduced from the top rate on ordinary income (38.6 percent) to the capital
gains tax rate (maximum of 15 percent). Advocates argued that this tax
cut would reduce the tax term and stimulate business investment.
51
The
Joint Committee on Taxation estimated that the dividend tax cut would
reduce revenue by more than $100 billion from 2003 to 2008.
52
Given this
high cost, it is worth assessing the impact of the cut. If the new view is cor-
rect, changes in dividend taxes have little or no impact on the cost of capital.
Table 10 reports results of our testing between the two views. In the
rst column we consider the relevance of dividend taxes by contrasting
the predictions of equations 5 and 6 in one specication. The difference is
Mihir A. Desai and Austan D. Goolsbee 317
51. See Hederman (2004) and Larry Kudlow, A Capital Idea from Microsoft,
National Review Online, posted on July 23, 2004 (www.nationalreview.com/kudlow/
kudlow200407230854.asp).
52. Joint Committee on Taxation (2003a).
Table 9. Regressions Testing Sensitivity of the Firm-Level Investment-Q
Relationship to Components of Q in Manufacturing and Nonmanufacturing
Industries
a
All Manufacturing Nonmanufacturing
Independent variable industries industries only industries only
q/(1 t) 0.0231 0.0210 0.0169
(0.0011) (0.0015) (0.0011)
Tax term, equipment 0.8895 1.1545 0.7034
(0.3173) (0.6943) (0.2853)
Tax term, structures 0.0169 0.0035 0.0307
(0.0452) (0.1762) (0.0405)
Ratio of cash ow to 0.0005 0.0001 0.0090
capital stock (0.0015) (0.0022) (0.0018)
No. of observations 141,629 68,680 111,059
Adjusted R
2
.376 .336 .381
Source: Authors regressions using data from Compustat.
a. The dependent variable in all regressions is the rms investment rate, as dened in the text, in a given year. All regressions
are as specied in the rst column of table 8. Numbers in parentheses are standard errors clustered at the rm level.
2581-04_Desai.qxd 1/18/05 13:28 Page 317
simply whether the q term is adjusted by the dividend tax preference
parameter or not. In the empirical specication of the rst column, we use
all years for which we have rm data. The measure of q that is interacted
with the term has a positive coefcient and is highly signicant.
The measure that is not interacted with the term is insignicant
and actually has a negative coefcient. The two coefcients are signi-
cantly different from one another as well. In other words, although the
marginal source of funds for these rms cannot be observed directly, their
investment behavior is consistent with their treating retained earnings as
the marginal source, and this implies that the new view is the correct one.
One major criticism of most previous analyses of the impact of divi-
dend tax rates on investment and other economic behavior has been that
changes in the rates themselves do not occur in isolation, but instead
accompany changes in the top marginal rate on ordinary income.
53
In the
later part of our sample, however (19972003), the tax changes that
1
1
_
,
c
1
1
_
,
c
_
,
t
Mihir A. Desai and Austan D. Goolsbee 319
54. We also tried using only the personal tax rates from Poterba (2004) to take out any
potential bias that the trends in corporate and nontaxable investor shares of dividends
received might have on average marginal tax rates. This made no difference to our results
and consistently showed evidence in favor of the new view. Note that our results are not
identied by the level of the dividend tax term (which gets absorbed in the year dummies)
but instead by the interaction with q.
55. Auerbach and Hassett (2002).
56. Carroll, Hassett, and Mackie (2003) simulate the impact in more detail under vari-
ous assumptions.
2581-04_Desai.qxd 1/18/05 13:28 Page 319
equivalent of dividing the cost of capital by 1.035. The equipment tax
term in 2003 was about 1.031, so this would have an effect on investment
of approximately the same magnitude as converting the tax code to com-
plete and immediate expensing of all equipment investment in 2003 (since
dividing the tax term by 1.035 would yield a value of approximately 1, the
same as immediate expensing). We will show in the next subsection,
however, that changes to the tax term of that magnitude may not increase
investment by much in the short run during this sample period.
The Impact of Partial Expensing
Although we nd no impact of the dividend tax cuts on investment, the
other tax incentives enacted during the early 2000s, specically the depre-
ciation allowance and partial expensing changes, directly reduced the tax
term under either view of the dividend tax and should have stimulated
investment. Their apparent failure to do so has led some to argue that tax
policy is not effective.
magnitude of the cuts. In 2002 President Bush signed a change in
the tax code to allow for partial expensing of equipment; this change was
made retroactive to cover all investment in 2002. In essence this rule
change broke an investment into two parts. Thirty percent of the invest-
ment is immediately expensed. The remaining 70 percent is depreciated
according to the normal schedule (which allows the rm to write off some
portion in the rst year, some portion in the second, and so on, for the tax
life of the asset). Given that a fairly large share of the investment not
being expensed already gets depreciated in the rst year, this new law
heavily weighted the depreciation allowances toward the rst year.
57
In
2003 the law was changed again (and again made retroactive to cover
investments made at any time during the year) to allow for rst-year
expensing of 50 percent of the investment. Although this provision was
scheduled to expire at the end of 2003, it was extended to 2004 and may
be extended further in the futureas of this writing, many legislators and
commentators are arguing that it should be made permanent.
320 Brookings Papers on Economic Activity, 2:2004
57. Cohen, Hassett, and Hansen (2002) provide a comprehensive analysis of the 2002
change.
2581-04_Desai.qxd 1/18/05 13:28 Page 320
These incentives were costly to provide, of course. The Joint Commit-
tee on Taxation estimated the cost of the changes in 2002 at about $35 bil-
lion and the cost of the higher expensing in 2003 and 2004 at about $32
billion and $53 billion, respectively.
58
Extending it indenitely would pre-
sumably entail similar annual costs.
To estimate the effect of the changed investment incentives, we com-
pute the increase in investment implied by the changes to depreciation
allowances. The last two columns of appendix table A-1 report the change
in the tax term from 2001 to 2003, averaged across industries at the three-
digit NAICS level. The rst column considers the overall change in the
tax cost, and the second the change in the tax term for equipment only.
Not surprisingly, the amounts differ across industries depending on the
nature of the investment goods they purchase. Airlines, for example,
invest mostly in equipment and mostly in long-lived assets such as air-
craft. Long-lived assets that qualify for bonus depreciation receive the
largest boost from allowing 50 percent immediate expensing (since they
were depreciated over a longer period before) and thus provide the largest
changes in the tax term. Firms in industries such as real estate and hotels
invest little in equipment, and what equipment they buy tends to consist
of computers and other short-lived assets for which immediate expensing
is not as large an improvement.
The table shows that even these rather dramatic changes to the depreci-
ation and expensing rules did not have a very large impact on the tax term.
The average change in the equipment tax term across all rms is about
0.03 (0.02 after incorporation of the equipment share). Such a change is
modest compared with changes such as the investment tax credit of 1962,
its restoration in the early 1970s, or the increases in the depreciation
allowance in 1981, all of which changed the overall tax term by around
0.10. Figure 5 depicts the industry-average equipment tax term over
roughly the last half century and shows that the most recent changes in
investment incentives have been modest by historical standards.
This relatively small effect stems from several factors. First, the value
of an acceleration in depreciation allowances is a function of the corporate
tax rate: with corporate tax rates already lower than they were in previous
decades, altering depreciation schedules has a more muted effect. Second,
the well-documented shift of investment toward computers and other
Mihir A. Desai and Austan D. Goolsbee 321
58. Joint Committee on Taxation (2002, 2003a, 2003b).
2581-04_Desai.qxd 1/18/05 13:28 Page 321
322 Brookings Papers on Economic Activity, 2:2004
equipment with shorter lives has meant that accelerated depreciation pro-
vides less relief. The average net present value of depreciation allowances
for equipment investment in 2001 was already approximately 90 percent
of the investment value even before the tax cuts, suggesting that even com-
plete expensing (raising the net present value to 1) would provide limited
additional benet. Given the smaller magnitude of the 2002 and 2003
cuts, it is unsurprising that such incentives could not overcome the dra-
matic drop in investment induced by the remarkable drop in q over the
period. Our estimates suggest that these incentives do work as they are
designed, but that their magnitude is simply too small to counteract the
aggregate trend.
how much did these incentives increase investment? To esti-
mate the precise impact of the tax changes on investment, we return to the
tax-adjusted q model. To use that model to simulate the impact of the tax
cuts in 2002 and 2003, we need to compute the transition path for invest-
ment in the standard q model.
59
Auerbach outlines a linearization that
59. See Abel (1981) and Summers (1981) for discussion.
0.95
1.00
1.05
1.10
1.15
1.20
1954 1958 1962 1966 1970 1974 1978 1982 1986 1990 1994 1998 2002
Source: Authors calculations based on Jorgenson and Yun (2001).
a. Unweighted average of q/(1 T) (1 )/(1 T) for 300 three-digit NAICS industries.
L
Figure 5. Aggregate Tax Term for Equipment,
a
19502003
2581-04_Desai.qxd 1/18/05 13:28 Page 322
Mihir A. Desai and Austan D. Goolsbee 323
makes this particularly easy,
60
and we adopt his notation to derive the pre-
dicted effects. Assuming a Cobb-Douglas production function with a capi-
tal share of 1 a, a real before-tax interest rate of r, quadratic adjustment
costs of (the reciprocal of the true coefcient on Q in our regressions),
and an adjustment cost-modied depreciation rate for capital in the rm
of
(the specic formula for which is given below), Auerbach shows that,
for an unanticipated permanent change in tax policy, the capital stock fol-
lows a simple partial adjustment model with
K
t
1
(K* K
t
), where K* is
the desired capital stock. The rate of adjustment
1
follows the formula
To compute this adjustment rate empirically, we assume a real interest
rate of 5 percent. We compute a, the complement of the capital share, as
1 minus the gross output share of value added for each industry as
reported in the disaggregated National Income and Product Accounts data
for 1998. We take the true coefficient on Q to be 1, following the results
above. We compute
. 7
4
2
1
2
+
+
( )
r r
a r
60. Auerbach (1989).
61. Following the traditional literature, we make these calculations assuming that the
elasticity of K* with respect to the cost of capital is equal to 1, for a Cobb-Douglas pro-
duction function for a given level of output. Such a gure is consistent with the empirical
ndings surveyed in Hassett and Hubbard (2002) but larger than the ndings discussed in
Chirinko (1993) or in Chirinko, Fazzari, and Meyer (1999). If labor were held xed rather
than output, our calculations would be scaled by 1/, where is the complement of the
industry capital share (averaging about .66 to .70 in our data). To compute the total effect
with varying output would require a full general equilibrium model for all sectors with
industry-specic demand information. More details on the assumptions behind such calcu-
lations can be found in Coen (1969) and Hall and Jorgenson (1969).
2581-04_Desai.qxd 1/18/05 13:28 Page 323
324 Brookings Papers on Economic Activity, 2:2004
To compute the effect of these policies over the past two years, we
assume that the depreciation changes were unanticipated and thought to
be permanent.
62
We first derive the optimal capital stock and amount of
adjustment in the first year (2002). We then calculate the new optimal
capital stock for 2003 (after the second tax cut) and the amount of
adjustment based on the new gap between K* and actual K (where
actual K is higher than it was in 2001 because of the investment under-
taken in 2002). Averaging for each three-digit industry and summing
over the two years, we estimate the impact of the tax cuts on the capital
stock for each industry (table 11). The average increase for the period is
only about 1.0 to 1.5 percent, and so it is immediately clear why these
tax cuts have seemed to have little success in stemming the investment
declines: their short-run stimulus effect is too small. This is not a refu-
tation of the view that taxes matter. The tax cuts were effective in
changing incentivesthey simply were not large enough to counteract
the double-digit declines in investment rates observed in the 2000s.
Changes to depreciation allowances simply do not have much impact
when the system is already so close to full expensing and when aggre-
gate declines in market value (and therefore in q) are so large. Since
firms are moving asymptotically to the optimal capital stock, the
effects of the policy change will be smaller in later years than in the
first two years. After 2004 the average total increase will still be less
than 2 percent.
Conclusion
This paper has addressed two major questions arising from the puz-
zling investment experience of the 2000s thus far: First, to what extent
was the equity bubble of the 1990s correlated with the decline in invest-
ment in the 2000s? And second, why didnt the major tax cuts of 200203
do more to restore investment to normal levels?
62. That the changes were unanticipated is probably fairly accurate. An assumption of
permanence seems reasonable, because although the changes were announced as tempo-
rary, from the moment they were passed many have been advocating that they be made per-
manent (and indeed the changes have already been extended to 2004). We assume
permanence here because it considerably simplies the computation of the investment path.
2581-04_Desai.qxd 1/18/05 13:28 Page 324
Mihir A. Desai and Austan D. Goolsbee 325
Table 11. Estimates of Change in Capital Stock Attributable to Tax Cuts of 2002
and 2003, by Industry
Change in capital stock,
200203
a
(percent)
Using tax term Using tax term
NAICS for equipment for equipment
code Industry and structures only
481 Air Transp. 1.58 1.68
514 Information Servs. & Data Processing Servs. 1.47 1.79
492 Couriers & Messengers 1.44 1.52
561 Administrative & Support Servs. 1.43 1.63
213 Support Activities for Mining 1.42 1.61
487 Scenic & Sightseeing Transp. 1.42 1.63
488 Support Activities for Transp. 1.42 1.63
485 Transit & Ground Passenger Transp. 1.39 1.52
541 Professional, Scientic, & Technical Services 1.32 1.58
233 Building, Developing, & General Contracting 1.32 1.38
234 Heavy Construction 1.32 1.38
235 Special Trade Contractors 1.32 1.38
313 Textile Mills 1.31 1.58
314 Textile Product Mills 1.30 1.60
316 Leather & Allied Product Mfg. 1.30 1.54
323 Printing & Related Support Activities 1.30 1.57
322 Paper Mfg. 1.28 1.46
421 Wholesale Trade, Durable Goods 1.27 1.58
422 Wholesale Trade, Nondurable Goods 1.27 1.58
493 Warehousing & Storage 1.26 1.63
513 Broadcasting & Telecommunications 1.25 1.68
511 Publishing 1.25 1.55
114 Fishing, Hunting, & Trapping 1.25 1.68
621 Ambulatory Health Care Servs. 1.24 1.61
331 Primary Metal Mfg. 1.22 1.43
523 Securities, Commodity Contracts, & Other Fin. 1.20 1.59
334 Computer & Electronic Product Mfg. 1.19 1.58
321 Wood Product Mfg. 1.18 1.53
336 Transp. Equip. Mfg. 1.18 1.46
315 Apparel Mfg. 1.18 1.56
484 Truck Transp. 1.16 1.36
312 Beverage & Tobacco Product Mfg. 1.15 1.39
337 Furniture & Related Product Mfg. 1.14 1.50
332 Fabricated Metal Product Mfg. 1.12 1.35
524 Insurance Carriers & Related Activities 1.12 1.41
333 Machinery Mfg. 1.12 1.43
327 Nonmetallic Mineral Product Mfg. 1.11 1.35
339 Miscellaneous Mfg. 1.11 1.41
811 Repair & Maintenance 1.11 1.45
311 Food Mfg. 1.10 1.42
335 Electrical Equip., Appliance, & Component Mfg. 1.09 1.37
(continued)
2581-04_Desai.qxd 1/18/05 13:28 Page 325
326 Brookings Papers on Economic Activity, 2:2004
326 Plastics & Rubber Products Mfg. 1.07 1.29
622 Hospitals 1.03 1.82
324 Petroleum & Coal Products Mfg. 1.02 1.18
212 Mining (except Oil & Gas) 0.97 1.37
812 Personal & Laundry Servs. 0.95 1.43
113 Forestry & Logging 0.94 1.20
512 Motion Picture & Sound Recording 0.93 1.66
722 Food Servs. & Drinking Places 0.88 1.48
532 Rental & Leasing Servs. 0.86 1.00
325 Chemical Mfg. 0.86 1.28
522 Credit Intermediation & Related Activities 0.80 1.28
624 Social Assistance 0.77 1.59
482 Rail Transp. 0.73 1.39
562 Waste Management & Remediation Servs. 0.71 1.44
486 Pipeline Transp. 0.66 1.75
813 Religious, Grantmaking, Civic, Prof. & Similar 0.65 1.60
533 Lessors of Nonn. Intangible Assets 0.63 1.10
111 Crop Production 0.61 0.70
444 Building Mtrl & Garden Equip. & Supplies 0.58 1.56
448 Clothing & Clothing Accessories Stores 0.58 1.56
443 Electronics & Appliance Stores 0.58 1.56
445 Food & Beverage Stores 0.58 1.56
442 Furniture & Home Furnishings Stores 0.58 1.56
447 Gasoline Stations 0.58 1.56
452 General Merchandise Stores 0.58 1.56
446 Health & Personal Care Stores 0.58 1.56
453 Miscellaneous Store Retailers 0.58 1.56
441 Motor Vehicle & Parts Dealers 0.58 1.56
454 Nonstore Retailers 0.58 1.56
451 Sporting Goods, Hobby, Book, & Music Stores 0.58 1.56
221 Utilities 0.56 1.36
623 Nursing & Residential Care Facilities 0.56 1.74
112 Animal Production 0.54 0.66
611 Educational Servs. 0.49 1.72
713 Amusement, Gambling, & Recreation 0.41 1.31
711 Performing Arts, Spectator Sports, & Related 0.38 1.54
211 Oil & Gas Extraction 0.26 1.48
525 Funds, Trusts, & Other Fin. Vehicles 0.18 1.46
721 Accommodation 0.18 1.46
531 Real Estate 0.03 0.72
Source: Authors calculations using data from Compustat as described in appendix A.
a. Change in capital stock associated with changes in the expensing provisions by industry as described in the text.
Table 11. Estimates of Change in Capital Stock Attributable to Tax Cuts of 2002
and 2003, by Industry (continued)
Change in capital stock,
200203
a
(percent)
Using tax term Using tax term
NAICS for equipment for equipment
code Industry and structures only
2581-04_Desai.qxd 1/18/05 13:28 Page 326
Mihir A. Desai and Austan D. Goolsbee 327
The data at the rm, asset, and industry level do not support the popu-
lar explanation of how capital overhang affected the investment market in
the 2000s. The general evidence shows that rapid growth of investment in
the 1990s had very little correlation with the investment declines in the
2000s. The evidence further indicates that the rm-level relationship
between investment and q has not changed noticeably in the recent period
for rms that saw large increases in their market value or that invested
heavily in the 1990s. Instead the rise and fall of equity prices, in the con-
text of a conventional tax-adjusted q model that better accounts for mea-
surement error in measuring marginal q, is the best explanation for the
investment experience of the recent past.
This conventional tax-adjusted q model then serves as the basis for our
analysis of the impact of the recent tax cuts and their seeming impotence.
Our results show that the dividend tax cut, despite its high revenue cost,
had minimal, if any, impact on marginal investment incentives. The
results strongly favor the new view of dividend taxation, in which such
taxes are capitalized into share prices and do not affect marginal incen-
tives. Similarly, the partial expensing provisions passed in 2002 and 2003
were not large enough to provide much counterweight to the declines in
aggregate investment. Our estimates suggest that tax policies contributed
to an increase in the capital stock of only 1 to 2 percent.
2581-04_Desai.qxd 1/18/05 13:28 Page 327
328 Brookings Papers on Economic Activity, 2:2004
AP P E NDI X A
Data Sources and Denitions
Firm-Level Financial Data
Annual data for all companies in the Compustat database, from 1950
on, are accessed through Wharton Research Data Services (WRDS).
Market Valuation of the Capital Stock
The Compustat series Property, Plant, and EquipmentTotal (Net)
is used as a measure of capital equipment; Capital Expenditures (State-
ment of Cash Flows) is used as a measure of capital expenditure. Each of
these measures is converted to constant dollars by dividing by the current
value of the producer price index (PPI) for capital goods, taken from the
website of the Bureau of Labor Statistics.
Three factors enter into the current valuation of the capital stock. The
rst is changes in the prices of capital goods held over from previous
years. Our conversion to constant dollars sidesteps this component. The
second is additions to capital through investment expenditure. The third is
depletion of capital on hand through depreciation.
The rms current real capital stock can be thought of as the sum of the
nondepreciated stocks of all previous years plus investment in the current
year. Following Cummins, Hassett, and Hubbard (1994), and assuming a
constant rate of depreciation , we calculate the current capital stock as
K
T
K
0
(1 )
T
+ I
1
(1 )
T 1
+ + I
T 1
(1 ) + I
T
.
For example, the rm starts in period 0 with capital stock K
0
, but only
the nondepreciated part of this stock, (1 )K
0
, remains to be carried over
to the next period. Some of this carried-over capital is used up in produc-
ing output in the second period, leaving (1 )
2
K
0
to be carried over to the
third period, and so forth. By period T, then, only (1 )
T
K
0
is carried
over from period 0. Similar reasoning explains the coefcients on the lev-
els of investment carried over to period T from all previous years. I
T
rep-
resents investment expenditure in period T.
Given the ending levels of the capital stock for all years, including the
nal year, and the nal years investment spending (all deated by the PPI
2581-04_Desai.qxd 1/18/05 13:28 Page 328
for capital goods), we can solve for the average rate of depreciation for
each rm. This average rate of depreciation is then applied sequentially,
from the rst observed year for each rm to the last, to derive an esti-
mated capital stock for each rm-year observation.
Conversion of Inventory from Book to Market Valuation
The Compustat series InventoriesTotal is used as a measure of
the current value of inventory holdings. As in Cummins, Hassett, and
Hubbard (1994), we convert inventory levels from their book value to
market value (on a last-in, first-out, or LIFO, basis) by adjusting the
lagged book value of carried-over inventories for year-to-year changes in
the prices of finished goods. How the adjustment mechanism is imple-
mented depends on whether final inventories increase or decrease from
one year to the next. If inventories increase, those goods carried over
from the previous year are revalued at current prices, as is the net addi-
tion to total inventories:
Inv
m
t
Inv
m
t 1
(P
t
/P
t 1
) + Inv
t
, if Inv
t
0.
Essentially, under LIFO valuation rules, the ending levels of invento-
ries include all goods that are carried over from the previous year plus
unsold current production. All inventories carried into the current year
remain at the end of the year and are revalued at current prices. The net
addition to inventories is already measured at current prices and so needs
no further adjustment.
On the other hand, if inventories decrease during the current year, it is
assumed that all current production has been sold as well as some part of
inventories carried over from the previous year. All goods remaining at
the end of the year are then valued at current prices:
Inv
m
t
(Inv
m
t 1
+ Inv
t
Inv
t 1
)(P
t
/P
t 1
)
(Inv
m
t 1
+ Inv
t
)(P
t
/P
t 1
), if Inv
t
< 0.
Operating Income
The Compustat series Operating Income before Depreciation and
Operating Income after Depreciation are used as measures of net
income. Each was converted from nominal to real terms by dividing by
Mihir A. Desai and Austan D. Goolsbee 329
2581-04_Desai.qxd 1/18/05 13:28 Page 329
330 Brookings Papers on Economic Activity, 2:2004
the PPI for nished goods, taken from the website of the Bureau of Labor
Statistics.
Analysts Earnings Estimates
Consensus analysts estimates of rms earnings per share in future
years were taken from the I/B/E/S summary statistics data maintained on
WRDS. The variables in this le include the number of estimates and the
mean, median, and standard deviation of estimates for a number of scal
periods (quarters or years) into the future. We merged the Compustat
rm-level nancial data with the I/B/E/S rm-level analysts estimates.
We used the summary estimate made during the latest month before the
end of the rms scal year.
Asset-Level Tax Term
Data for the asset-level tax term come from Dale Jorgenson of Harvard
University; his methodology is described in Jorgenson and Yun (2001).
These data provide, for each asset type, an estimate of the net present
value of depreciation allowances z, the investment tax credit rate, and the
depreciation rate, as well as the capital stock and the average corporate
tax rate. We compute as ITC + tz and the full tax term as (1 ITC
tz)/(1 t). The calculations are myopic in that they do not include the
impact of expected future tax changes; current tax rates are assumed to be
permanent. We modify the net present values of depreciation allowances
in 2002 and 2003, to account for the changes in the partial expensing
rules. We recomputed z, for 2002, using a 70-30 weighted average of the
old z and 1; we do the same, but with 50-50 weights, for 2003.
Industry- and Firm-Level Tax Terms
To derive industry-level values of the tax term for equipment and
structures as well as to derive industry-level depreciation rates, we use the
1997 capital ow tables of the BEA and compute the share of equipment
and structures investment by asset type for each industry at approximately
2581-04_Desai.qxd 1/18/05 13:28 Page 330
Mihir A. Desai and Austan D. Goolsbee 331
the three-digit NAICS level. We match these weights to the Jorgenson tax
term gures by year for each asset type to compute a weighted-average
tax term in each year for each industry. We then merge that series to each
rm-year based on its rst listed NAICS code in Compustat (table A-1).
Present Value of Depreciation Allowances on Past Investment
To estimate the value of A, the net present value (NPV) of deprecia-
tion allowances on past investments, we sort firms according to the
weighted average of depreciation rates on the types of equipment in
which firms in their industry invest. Using the inverse of this average
depreciation rate as an estimate of the lifetime of the firms capital, we
assume that all firms in the industry have a discount rate of 10 percent
and use double-declining-balance depreciation until straight-line depre-
ciation exceeds it, and then switch to straight-line. We then multiply the
NPV of the remaining depreciation allowances on investment from a
given year in the assets life by the investment-to-capital ratio lagged
that many periods. For example, if the actual depreciation allowances for
a three-year-lived good costing $1 were one-third each year (pure
straight-line depreciation), then the NPV of the allowances in the year of
the investment would be , the NPV of the
allowances remaining one year later would be , and
the NPV of allowances after two years would be z
age2
(.333). We would
then compute the value of depreciation on previous investments
as . Note that the NPV of depreciation
allowances for current (time t) investment is not included in this measure
(although it is in z); hence the computation of A for an industry whose
asset life is three years has only two terms.
We compute the NPV assuming an asset life of three years for any rm
for which the inverse of its average depreciation rate is between 3 and 4,
four years for any rm for which the inverse is between 4 and 5, and so
on, but with a cap at nine years (a few rms had average equipment lives
of slightly over ten years).
A t z
I
K
z
I
K
age
t
age
t
_
,
+
_
,
1
]
1
1
1
2
2
z
age
+
_
,
1
333
333
1 1
.
.
.
z
age
+ +
_
,
0
2
333
333
1 1
333
1 1
.
.
.
.
.
2581-04_Desai.qxd 1/18/05 13:28 Page 331
T
a
b
l
e
A
-
1
.
E
s
t
i
m
a
t
e
s
o
f
A
d
j
u
s
t
m
e
n
t
s
t
o
T
a
x
C
h
a
n
g
e
s
b
y
I
n
d
u
s
t
r
y
E
q
u
i
p
m
e
n
t
E
q
u
i
p
m
e
n
t
s
h
a
r
e
o
f
t
o
t
a
l
E
q
u
i
p
m
e
n
t
E
q
u
i
p
m
e
n
t
N
A
I
C
S
a
n
d
E
q
u
i
p
m
e
n
t
i
n
v
e
s
t
m
e
n
t
a
n
d
E
q
u
i
p
m
e
n
t
a
n
d
E
q
u
i
p
m
e
n
t
c
o
d
e
I
n
d
u
s
t
r
y
s
t
r
u
c
t
u
r
e
s
o
n
l
y
(
p
e
r
c
e
n
t
)
s
t
r
u
c
t
u
r
e
s
o
n
l
y
s
t
r
u
c
t
u
r
e
s
o
n
l
y
1
1
1
C
r
o
p
P
r
o
d
u
c
t
i
o
n
0
.
1
4
8
0
.
1
4
3
9
1
0
.
1
5
4
0
.
1
4
3
0
.
0
2
8
0
.
0
3
1
1
1
2
A
n
i
m
a
l
P
r
o
d
u
c
t
i
o
n
0
.
1
5
4
0
.
1
4
7
8
7
0
.
1
7
2
0
.
1
5
3
0
.
0
2
4
0
.
0
2
9
1
1
3
F
o
r
e
s
t
r
y
&
L
o
g
g
i
n
g
0
.
3
1
8
0
.
3
0
4
8
5
0
.
1
6
4
0
.
1
4
5
0
.
0
2
2
0
.
0
2
7
1
1
4
F
i
s
h
i
n
g
,
H
u
n
t
i
n
g
,
&
T
r
a
p
p
i
n
g
0
.
2
7
2
0
.
2
6
2
8
1
0
.
1
0
6
0
.
0
9
4
0
.
0
3
3
0
.
0
4
3
2
1
1
O
i
l
&
G
a
s
E
x
t
r
a
c
t
i
o
n
0
.
3
1
5
0
.
2
5
2
2
1
0
.
1
7
2
0
.
0
9
1
0
.
0
0
7
0
.
0
3
3
2
1
2
M
i
n
i
n
g
(
e
x
c
e
p
t
O
i
l
&
G
a
s
)
0
.
3
0
0
0
.
2
8
6
8
2
0
.
1
3
5
0
.
1
1
8
0
.
0
2
3
0
.
0
3
0
2
1
3
S
u
p
p
o
r
t
f
o
r
M
i
n
i
n
g
0
.
3
7
9
0
.
3
7
1
9
0
0
.
1
7
1
0
.
1
6
1
0
.
0
2
8
0
.
0
3
1
2
2
1
U
t
i
l
i
t
i
e
s
0
.
2
4
0
0
.
2
0
7
5
8
0
.
1
4
0
0
.
0
9
2
0
.
0
1
9
0
.
0
3
8
2
3
3
B
u
i
l
d
i
n
g
,
D
e
v
e
l
o
p
i
n
g
,
&
G
e
n
.
C
o
n
t
r
a
c
t
i
n
g
0
.
3
6
5
0
.
3
6
2
9
7
0
.
1
8
7
0
.
1
8
2
0
.
0
2
6
0
.
0
2
7
2
3
4
H
e
a
v
y
C
o
n
s
t
r
u
c
t
i
o
n
0
.
3
6
5
0
.
3
6
2
9
7
0
.
1
8
7
0
.
1
8
2
0
.
0
2
6
0
.
0
2
7
2
3
5
S
p
e
c
i
a
l
T
r
a
d
e
C
o
n
t
r
a
c
t
o
r
s
0
.
3
6
5
0
.
3
6
2
9
7
0
.
1
8
7
0
.
1
8
2
0
.
0
2
6
0
.
0
2
7
3
1
1
F
o
o
d
M
f
g
.
0
.
3
1
1
0
.
2
9
6
8
4
0
.
1
5
5
0
.
1
3
5
0
.
0
2
6
0
.
0
3
2
3
1
2
B
e
v
e
r
a
g
e
&
T
o
b
a
c
c
o
P
r
o
d
u
c
t
M
f
g
.
0
.
3
1
6
0
.
3
0
3
8
7
0
.
1
6
1
0
.
1
4
4
0
.
0
2
6
0
.
0
3
1
3
1
3
T
e
x
t
i
l
e
M
i
l
l
s
0
.
3
3
5
0
.
3
2
4
8
8
0
.
1
2
8
0
.
1
1
7
0
.
0
2
9
0
.
0
3
4
3
1
4
T
e
x
t
i
l
e
P
r
o
d
u
c
t
M
i
l
l
s
0
.
3
6
3
0
.
3
4
8
8
7
0
.
1
6
1
0
.
1
4
4
0
.
0
2
7
0
.
0
3
2
3
1
5
A
p
p
a
r
e
l
M
f
g
.
0
.
3
4
2
0
.
3
2
4
8
3
0
.
1
5
3
0
.
1
3
2
0
.
0
2
6
0
.
0
3
3
3
1
6
L
e
a
t
h
e
r
&
A
l
l
i
e
d
P
r
o
d
u
c
t
M
f
g
.
0
.
3
4
4
0
.
3
3
2
8
9
0
.
1
5
6
0
.
1
4
2
0
.
0
2
8
0
.
0
3
2
3
2
1
W
o
o
d
P
r
o
d
u
c
t
M
f
g
.
0
.
3
3
4
0
.
3
1
9
8
4
0
.
1
4
2
0
.
1
2
4
0
.
0
2
6
0
.
0
3
3
3
2
2
P
a
p
e
r
M
f
g
.
0
.
3
0
7
0
.
3
0
1
9
3
0
.
1
3
5
0
.
1
2
8
0
.
0
3
0
0
.
0
3
3
3
2
3
P
r
i
n
t
i
n
g
&
R
e
l
a
t
e
d
S
u
p
p
o
r
t
0
.
3
5
6
0
.
3
4
3
8
8
0
.
1
6
3
0
.
1
4
7
0
.
0
2
7
0
.
0
3
2
3
2
4
P
e
t
r
o
l
e
u
m
&
C
o
a
l
P
r
o
d
u
c
t
s
M
f
g
.
0
.
2
4
3
0
.
2
3
7
9
1
0
.
1
5
5
0
.
1
4
4
0
.
0
3
0
0
.
0
3
3
3
2
5
C
h
e
m
i
c
a
l
M
f
g
.
0
.
2
8
6
0
.
2
6
8
8
0
0
.
1
8
0
0
.
1
5
0
0
.
0
2
2
0
.
0
3
0
3
2
6
P
l
a
s
t
i
c
s
&
R
u
b
b
e
r
P
r
o
d
u
c
t
s
M
f
g
.
0
.
3
0
0
0
.
2
9
0
8
8
0
.
1
3
3
0
.
1
2
1
0
.
0
2
6
0
.
0
3
0
3
2
7
N
o
n
m
e
t
a
l
l
i
c
M
i
n
e
r
a
l
P
r
o
d
u
c
t
M
f
g
.
0
.
3
0
2
0
.
2
9
1
8
8
0
.
1
5
4
0
.
1
3
9
0
.
0
2
7
0
.
0
3
1
3
3
1
P
r
i
m
a
r
y
M
e
t
a
l
M
f
g
.
0
.
3
4
5
0
.
3
3
2
8
8
0
.
1
5
6
0
.
1
4
1
0
.
0
2
6
0
.
0
3
0
A
d
j
u
s
t
m
e
n
t
r
a
t
e
(
1
)
D
e
p
r
e
c
i
a
t
i
o
n
r
a
t
e
(
)
C
h
a
n
g
e
i
n
t
a
x
t
e
r
m
,
2
0
0
1
0
3
2581-04_Desai.qxd 1/18/05 13:28 Page 332
3
3
2
F
a
b
r
i
c
a
t
e
d
M
e
t
a
l
P
r
o
d
u
c
t
M
f
g
.
0
.
3
3
9
0
.
3
2
8
8
9
0
.
1
6
2
0
.
1
4
7
0
.
0
2
4
0
.
0
2
9
3
3
3
M
a
c
h
i
n
e
r
y
M
f
g
.
0
.
3
7
1
0
.
3
5
3
8
5
0
.
1
9
6
0
.
1
7
1
0
.
0
2
3
0
.
0
2
9
3
3
4
C
o
m
p
u
t
e
r
&
E
l
e
c
t
r
o
n
i
c
P
r
o
d
u
c
t
M
f
g
.
0
.
3
9
5
0
.
3
7
7
8
6
0
.
2
1
2
0
.
1
8
6
0
.
0
2
4
0
.
0
3
0
3
3
5
E
l
e
c
t
.
E
q
u
i
p
.
,
A
p
p
l
i
a
n
c
e
,
&
C
o
m
p
o
n
e
n
t
M
f
g
.
0
.
3
2
7
0
.
3
1
2
8
6
0
.
1
6
6
0
.
1
4
7
0
.
0
2
5
0
.
0
3
0
3
3
6
T
r
a
n
s
p
.
E
q
u
i
p
.
M
f
g
.
0
.
3
6
6
0
.
3
5
3
8
8
0
.
1
7
1
0
.
1
5
5
0
.
0
2
4
0
.
0
2
9
3
3
7
F
u
r
n
i
t
u
r
e
&
R
e
l
a
t
e
d
P
r
o
d
u
c
t
M
f
g
.
0
.
3
4
5
0
.
3
2
7
8
3
0
.
1
5
2
0
.
1
3
1
0
.
0
2
5
0
.
0
3
1
3
3
9
M
i
s
c
e
l
l
a
n
e
o
u
s
M
f
g
.
0
.
3
3
4
0
.
3
2
0
8
6
0
.
1
6
2
0
.
1
4
4
0
.
0
2
4
0
.
0
2
9
4
2
1
W
h
o
l
e
s
a
l
e
T
r
a
d
e
,
D
u
r
a
b
l
e
G
o
o
d
s
0
.
4
0
2
0
.
3
8
3
8
6
0
.
2
0
8
0
.
1
8
3
0
.
0
2
4
0
.
0
2
9
4
2
2
W
h
o
l
e
s
a
l
e
T
r
a
d
e
,
N
o
n
d
u
r
a
b
l
e
G
o
o
d
s
0
.
4
0
2
0
.
3
8
3
8
6
0
.
2
0
8
0
.
1
8
3
0
.
0
2
4
0
.
0
2
9
4
4
1
M
o
t
o
r
V
e
h
i
c
l
e
&
P
a
r
t
s
D
e
a
l
e
r
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
2
F
u
r
n
i
t
u
r
e
&
H
o
m
e
F
u
r
n
i
s
h
i
n
g
s
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
3
E
l
e
c
t
r
o
n
i
c
s
&
A
p
p
l
i
a
n
c
e
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
4
B
u
i
l
d
i
n
g
M
t
r
l
,
G
a
r
d
e
n
S
u
p
p
l
i
e
s
D
e
a
l
e
r
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
5
F
o
o
d
&
B
e
v
e
r
a
g
e
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
6
H
e
a
l
t
h
&
P
e
r
s
o
n
a
l
C
a
r
e
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
7
G
a
s
o
l
i
n
e
S
t
a
t
i
o
n
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
4
8
C
l
o
t
h
i
n
g
&
C
l
o
t
h
i
n
g
A
c
c
e
s
s
o
r
i
e
s
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
5
1
S
p
o
r
t
i
n
g
G
o
o
d
s
,
B
o
o
k
,
&
M
u
s
i
c
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
5
2
G
e
n
e
r
a
l
M
e
r
c
h
a
n
d
i
s
e
S
t
o
r
e
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
5
3
M
i
s
c
e
l
l
a
n
e
o
u
s
S
t
o
r
e
R
e
t
a
i
l
e
r
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
5
4
N
o
n
s
t
o
r
e
R
e
t
a
i
l
e
r
s
0
.
3
9
7
0
.
3
2
0
4
8
0
.
1
9
8
0
.
1
0
8
0
.
0
1
3
0
.
0
2
9
4
8
1
A
i
r
T
r
a
n
s
p
.
0
.
3
1
3
0
.
3
0
9
9
6
0
.
1
2
4
0
.
1
2
1
0
.
0
3
6
0
.
0
3
8
4
8
2
R
a
i
l
T
r
a
n
s
p
.
0
.
2
8
8
0
.
2
5
3
5
9
0
.
0
9
7
0
.
0
6
4
0
.
0
2
0
0
.
0
3
4
4
8
4
T
r
u
c
k
T
r
a
n
s
p
.
0
.
3
5
0
0
.
3
3
9
9
0
0
.
1
8
4
0
.
1
6
9
0
.
0
2
5
0
.
0
2
8
4
8
5
T
r
a
n
s
i
t
&
G
r
o
u
n
d
P
a
s
s
e
n
g
e
r
T
r
a
n
s
p
.
0
.
3
6
7
0
.
3
6
0
9
4
0
.
1
9
2
0
.
1
8
2
0
.
0
2
8
0
.
0
3
0
4
8
6
P
i
p
e
l
i
n
e
T
r
a
n
s
p
.
0
.
3
3
6
0
.
2
7
3
4
8
0
.
1
7
7
0
.
0
9
8
0
.
0
1
7
0
.
0
3
7
4
8
7
S
c
e
n
i
c
&
S
i
g
h
t
s
e
e
i
n
g
T
r
a
n
s
p
.
0
.
3
9
1
0
.
3
7
7
9
0
0
.
1
9
3
0
.
1
7
6
0
.
0
2
7
0
.
0
3
1
4
8
8
S
u
p
p
o
r
t
f
o
r
T
r
a
n
s
p
.
0
.
3
9
1
0
.
3
7
7
9
0
0
.
1
9
3
0
.
1
7
6
0
.
0
2
7
0
.
0
3
1
4
9
2
C
o
u
r
i
e
r
s
&
M
e
s
s
e
n
g
e
r
s
0
.
4
0
2
0
.
3
9
7
9
6
0
.
2
0
9
0
.
2
0
2
0
.
0
2
7
0
.
0
2
8
4
9
3
W
a
r
e
h
o
u
s
i
n
g
&
S
t
o
r
a
g
e
0
.
3
8
3
0
.
3
6
4
8
4
0
.
1
8
0
0
.
1
5
6
0
.
0
2
5
0
.
0
3
1
5
1
1
P
u
b
l
i
s
h
i
n
g
I
n
d
u
s
.
0
.
4
0
0
0
.
3
8
1
8
6
0
.
2
7
2
0
.
2
3
8
0
.
0
2
4
0
.
0
2
9
5
1
2
M
o
t
i
o
n
P
i
c
t
u
r
e
&
S
o
u
n
d
R
e
c
o
r
d
i
n
g
0
.
3
8
6
0
.
3
4
2
6
7
0
.
1
9
6
0
.
1
4
0
0
.
0
1
9
0
.
0
3
1
(
c
o
n
t
i
n
u
e
d
)
2581-04_Desai.qxd 1/18/05 13:28 Page 333
5
1
3
B
r
o
a
d
c
a
s
t
i
n
g
&
T
e
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
s
0
.
2
7
8
0
.
2
4
6
6
5
0
.
1
4
6
0
.
1
0
3
0
.
0
3
2
0
.
0
4
1
5
1
4
I
n
f
o
S
e
r
v
s
.
&
D
a
t
a
P
r
o
c
e
s
s
i
n
g
S
e
r
v
s
.
0
.
4
4
8
0
.
4
2
8
8
7
0
.
2
5
8
0
.
2
2
8
0
.
0
2
6
0
.
0
3
0
5
2
2
C
r
e
d
i
t
I
n
t
e
r
m
e
d
i
a
t
i
o
n
&
R
e
l
a
t
e
d
0
.
3
4
7
0
.
3
1
1
7
2
0
.
2
7
6
0
.
2
0
6
0
.
0
1
8
0
.
0
2
6
5
2
3
S
e
c
u
r
i
t
i
e
s
,
C
o
m
m
o
d
i
t
y
C
o
n
t
r
a
c
t
s
,
&
O
t
h
e
r
0
.
4
3
9
0
.
4
1
1
8
2
0
.
2
7
2
0
.
2
2
7
0
.
0
2
2
0
.
0
2
7
5
2
4
I
n
s
u
r
a
n
c
e
C
a
r
r
i
e
r
s
&
R
e
l
a
t
e
d
0
.
3
8
9
0
.
3
6
8
8
5
0
.
2
7
0
0
.
2
3
3
0
.
0
2
2
0
.
0
2
7
5
2
5
F
u
n
d
s
,
T
r
u
s
t
s
,
&
O
t
h
e
r
F
i
n
a
n
c
i
a
l
V
e
h
i
c
l
e
s
0
.
4
1
9
0
.
2
6
0
1
9
0
.
2
9
6
0
.
0
7
7
0
.
0
0
5
0
.
0
2
6
5
3
1
R
e
a
l
E
s
t
a
t
e
0
.
1
8
2
0
.
0
9
7
5
0
.
1
9
3
0
.
0
2
6
0
.
0
0
2
0
.
0
2
7
5
3
2
R
e
n
t
a
l
&
L
e
a
s
i
n
g
S
e
r
v
s
.
0
.
2
2
7
0
.
2
1
8
8
8
0
.
1
9
4
0
.
1
7
3
0
.
0
2
6
0
.
0
3
0
5
3
3
L
e
s
s
o
r
s
o
f
N
o
n
n
a
n
c
i
a
l
I
n
t
a
n
g
i
b
l
e
A
s
s
e
t
s
0
.
2
5
0
0
.
2
2
0
6
8
0
.
2
4
9
0
.
1
7
8
0
.
0
2
0
0
.
0
3
0
5
4
1
P
r
o
f
e
s
s
i
o
n
a
l
,
S
c
i
e
n
t
i
c
,
&
T
e
c
h
n
i
c
a
l
S
e
r
v
s
.
0
.
4
0
4
0
.
3
8
5
8
6
0
.
2
4
9
0
.
2
1
8
0
.
0
2
5
0
.
0
2
8
5
6
1
A
d
m
i
n
i
s
t
r
a
t
i
v
e
&
S
u
p
p
o
r
t
S
e
r
v
s
.
0
.
4
1
2
0
.
4
0
0
9
1
0
.
2
0
9
0
.
1
9
3
0
.
0
2
6
0
.
0
2
9
5
6
2
W
a
s
t
e
M
a
n
a
g
e
m
e
n
t
&
R
e
m
e
d
i
a
t
i
o
n
S
e
r
v
s
.
0
.
3
5
2
0
.
3
0
7
6
1
0
.
1
9
0
0
.
1
3
1
0
.
0
1
6
0
.
0
2
9
6
1
1
E
d
u
c
a
t
i
o
n
a
l
S
e
r
v
s
.
0
.
4
4
7
0
.
3
4
0
4
1
0
.
2
1
4
0
.
0
9
9
0
.
0
1
0
0
.
0
2
9
6
2
1
A
m
b
u
l
a
t
o
r
y
H
e
a
l
t
h
C
a
r
e
S
e
r
v
s
.
0
.
3
6
7
0
.
3
4
7
8
3
0
.
1
8
1
0
.
1
5
4
0
.
0
2
6
0
.
0
3
2
6
2
2
H
o
s
p
i
t
a
l
s
0
.
4
0
9
0
.
3
6
3
6
8
0
.
1
6
6
0
.
1
1
9
0
.
0
2
1
0
.
0
3
3
6
2
3
N
u
r
s
i
n
g
&
R
e
s
i
d
e
n
t
i
a
l
C
a
r
e
F
a
c
i
l
i
t
i
e
s
0
.
4
1
4
0
.
3
2
8
4
4
0
.
1
7
2
0
.
0
8
7
0
.
0
1
2
0
.
0
3
1
6
2
4
S
o
c
i
a
l
A
s
s
i
s
t
a
n
c
e
0
.
4
1
9
0
.
3
5
7
5
9
0
.
2
0
6
0
.
1
3
2
0
.
0
1
6
0
.
0
2
8
7
1
1
P
e
r
f
o
r
m
i
n
g
A
r
t
s
,
S
p
e
c
t
a
t
o
r
S
p
o
r
t
s
0
.
3
6
5
0
.
2
7
3
3
6
0
.
1
9
2
0
.
0
8
4
0
.
0
1
0
0
.
0
3
1
7
1
3
A
m
u
s
e
m
e
n
t
,
G
a
m
b
l
i
n
g
,
&
R
e
c
r
e
a
t
i
o
n
0
.
3
4
6
0
.
2
7
7
4
3
0
.
1
6
5
0
.
0
8
7
0
.
0
1
0
0
.
0
2
7
7
2
1
A
c
c
o
m
m
o
d
a
t
i
o
n
0
.
3
3
4
0
.
2
3
2
2
0
0
.
1
6
0
0
.
0
5
2
0
.
0
0
5
0
.
0
3
1
7
2
2
F
o
o
d
S
e
r
v
s
.
&
D
r
i
n
k
i
n
g
P
l
a
c
e
s
0
.
3
8
2
0
.
3
4
3
6
9
0
.
1
7
1
0
.
1
2
6
0
.
0
1
8
0
.
0
2
8
8
1
1
R
e
p
a
i
r
&
M
a
i
n
t
e
n
a
n
c
e
0
.
3
6
8
0
.
3
4
2
7
8
0
.
1
7
5
0
.
1
4
2
0
.
0
2
3
0
.
0
2
8
8
1
2
P
e
r
s
o
n
a
l
&
L
a
u
n
d
r
y
S
e
r
v
s
.
0
.
3
8
6
0
.
3
5
3
7
5
0
.
1
9
9
0
.
1
5
5
0
.
0
1
9
0
.
0
2
7
8
1
3
R
e
l
i
g
i
o
u
s
,
G
r
a
n
t
m
a
k
i
n
g
,
C
i
v
i
c
,
P
r
o
f
e
s
s
.
0
.
4
0
2
0
.
3
2
9
5
2
0
.
2
2
2
0
.
1
2
8
0
.
0
1
4
0
.
0
2
9
S
o
u
r
c
e
:
A
u
t
h
o
r
s
c
a
l
c
u
l
a
t
i
o
n
s
u
s
i
n
g
d
a
t
a
f
r
o
m
C
o
m
p
u
s
t
a
t
.
T
a
b
l
e
A
-
1
.
E
s
t
i
m
a
t
e
s
o
f
A
d
j
u
s
t
m
e
n
t
s
t
o
T
a
x
C
h
a
n
g
e
s
b
y
I
n
d
u
s
t
r
y
(
c
o
n
t
i
n
u
e
d
)
E
q
u
i
p
m
e
n
t
E
q
u
i
p
m
e
n
t
s
h
a
r
e
o
f
t
o
t
a
l
E
q
u
i
p
m
e
n
t
E
q
u
i
p
m
e
n
t
N
A
I
C
S
a
n
d
E
q
u
i
p
m
e
n
t
i
n
v
e
s
t
m
e
n
t
a
n
d
E
q
u
i
p
m
e
n
t
a
n
d
E
q
u
i
p
m
e
n
t
c
o
d
e
I
n
d
u
s
t
r
y
s
t
r
u
c
t
u
r
e
s
o
n
l
y
(
p
e
r
c
e
n
t
)
s
t
r
u
c
t
u
r
e
s
o
n
l
y
s
t
r
u
c
t
u
r
e
s
o
n
l
y
A
d
j
u
s
t
m
e
n
t
r
a
t
e
(
1
)
D
e
p
r
e
c
i
a
t
i
o
n
r
a
t
e
(
)
C
h
a
n
g
e
i
n
t
a
x
t
e
r
m
,
2
0
0
1
0
3
2581-04_Desai.qxd 1/18/05 13:28 Page 334
Mihir A. Desai and Austan D. Goolsbee 335
Note that our measure is an approximation because it assumes that tax
law remains unchanged over the whole sample period. In other words, the
NPV of depreciation allowances on current investment, z, that we get
from Jorgenson varies over time, but we do not have the entire deprecia-
tion schedules on which each z is based, and so we cannot let the calcula-
tion vary for A. We tried many different ways of computing A, for
example adopting different assumptions about depreciation methods, dif-
ferent discount rates, and so on, and found they had negligible impact on
the regression results.
AP P E NDI X B
Tax-Adjusted q with Dividend Taxes
WE BEGIN BY establishing the equilibrium condition that shareholders
receive their required return, r, from holding equity that provides taxable
dividends and capital gains, so that
(B1) rV
t
(1 )D
t
+ (1 c){E
t
[V
t + 1
] V
t
V
N
t
},
where is the tax rate on dividends and c is the accrual-equivalent tax rate
on capital gains. D
t
denotes dividends paid to shareholders in period t, V is
equity value, and V
N
t
denotes equity contributions made in period t. Given
that dividends and capital gains are alternative forms of returns to share-
holders, it is useful to summarize the relative tax penalty on dividends and
capital gains with the dividend tax preference parameter :
(B2) (1 )/(1 c).
Given that capital gains taxes are paid only when the gain is realized, is
considered to be less than 1.
63
Solving equation B1 forward, and imposing
the transversality condition that rm value cannot be innite in a nite
period, provides a value equation for the rm that implies
(B3) V E D V
t t t
N
t
0 0
0
( )
,
63. Even with similar rates on dividends and realized capital gains, < 1 is thought to
hold. Typically, the accrual-equivalent c is usually taken as one-quarter of the statutory rate
applicable to capital gains.
2581-04_Desai.qxd 1/18/05 13:28 Page 335
where is the appropriate after-tax discount factor. Equation B3 cor-
responds to the straightforward intuition that rm value at time 0 is the
present-discounted, tax-adjusted value of all future dividends, taking into
account any equity contributions required to maintain a proportional
shareholding in the rm.
Firm value maximization is subject to several constraints. Dividends
and equity issuance are constrained to be nonnegative.
64
The rms capi-
tal stock K evolves according to
(B4) K
t
K
t 1
(1 ) + I
t
,
where is a constant proportional rate of decay and I is investment. The
underlying cash ow identity for the rm is given by
(B5) (1 )[F(K
t
, L
t
) w
t
L
t
C(I
t
, K
t1
)p
t
]
+ V
N
t
+ A
t
D
t
+ I
t
p
t
(1
t
),
where is the statutory corporate tax rate, F(K, L) is rm output, L is
labor, w is the wage rate, C(I
t
, K
t 1
) is an adjustment cost function for
investment, and A captures the tax value of depreciation allowances on
previous investments. Variable p is the price of capital goods relative to
output, and is a summary measure of tax provisions that directly inu-
ence investment, such as the tax value of depreciation allowances and
investment tax credits. The source and measurement of are described in
appendix A. In short, equation B5 states that a rms after-tax cash ow
and its new equity issuances are sources of funds, which are used for
investment and for paying dividends, and the ps ensure that all terms are
properly price adjusted.
65
Given the expression for rm value in equation B3 and the constraints
discussed above, rm value maximization employs the following Hamil-
tonian equation:
(B6) H
t
D
t
V
N
t
1
t
[K
t
K
t 1
(1 ) I
t
]
2
t
D
t
3
t
V
N
t
.
336 Brookings Papers on Economic Activity, 2:2004
64. As Poterba and Summers (1985) note, repurchases can be allowed without loss of
generality. However, negative new equity issuances must be bounded by some maximum
amount, an assumption justied by the IRSs ability to characterize large, regularized
repurchases as dividends.
65. We abstract from debt and the presence of tax-deductible interest without loss of
generality.
2581-04_Desai.qxd 1/18/05 13:28 Page 336
In this setting,
1
t
,
2
t
, and
3
t
correspond to the shadow values of
capital goods, dividends, and negative equity issuances, respectively.
Substituting the value of dividends from the cash flow identity in equa-
tion B5, we can rewrite equation B6 as
(B7) H
t
(
2
t
){(1 )[F(K, L) wL C(I
t
, K
t 1
)p
t
]
+ V
N
t
+ I
t
p
t
(1 ) + A
t
} V
N
t
1
t
[K
t
K
t 1
(1 d) I
t
]
3
t
V
N
t .
Differentiating this Hamiltonian provides the relevant rst-order condi-
tions. The rst-order condition for investment is provided by
(B8) (
2
t
)(1 )C
I
p
t
p
t
(1 ) +
1
t
0,
and the conditions for dividends and net equity issuance are provided by
(B9) D
t
0;
2
t
0; and D
t
2
t
0
and
(B10) V
N
t
0; (
2
t
1
3
t
) 0; and V
N
t
(
2
t
1
3
t
)
0, respectively.
Rearranging the investment rst-order condition provided in equation
B8 provides an expression for q that corresponds to the shadow price for
capital:
(B11) (
1
t
/p
t
) q
t
(
2
t
)[(1 ) C
I
+ (1 )],
where C
I
is the marginal adjustment cost of new investment. In order to
put this in more familiar terms, we specify a conventional, quadratic
adjustment cost function:
(B12) C(I
t
, K
t 1
) (/2)[(I
t
/K
t 1
) ]
2
K
t 1
,
where is the adjustment cost parameter and represents an average
investment rate. This quadratic adjustment cost function allows us to rep-
resent equation B11 in more familiar terms. Differentiating the cost func-
tion with respect to I and substituting it into equation B11 yields
(B13)
I
K
q
t
t
t
t
_
,
( )
_
,
1
2
1 1
1
.
Mihir A. Desai and Austan D. Goolsbee 337
2581-04_Desai.qxd 1/18/05 13:29 Page 337
Equation B13 is the basic estimating equation commonly used in
the q-theory literature, with the slight peculiarity that q
t
is divided
by (
2
t
). In the existing literature and our discussion below,
is also referred to as Q rather than q. It will be important
in our discussion of measurement error to note that Q is actually com-
posed of two parts, associated with investment opportunities and taxes.
In order to consider under what conditions the additional, peculiar term
disappears, it is critical to specify the marginal source of nancing. To do
so we return to the conditions B9 and B10 and consider the alternative
cases where the marginal source of nancing is either retained earnings or
new equity issuance.
First, consider the case where the marginal source of nance is new
equity issuances. In this case,
3
t
0 and, consequently,
2
t
1, as indi-
cated by equation B10. In this case equation B13 becomes its more famil-
iar variant:
Now consider the alternative case where the marginal source of nance
is retained earnings rather than new equity issuance. This implies that div-
idends are positive and that
2
t
0. In turn, this implies that equation B13
can be rewritten as
Equations B14 and B15 provide alternative q-theory specications for
investment that incorporate different assumptions about the marginal
source of nance and, consequently, about the role of dividend taxation in
inuencing investment.
( ) . B15
1
1
1
1
1
1
1
I
K
q
c
t
t
t
_
,
_
,
1
]
1
1
1
_
,
(B14)
I
K
q
t
t
t
_
,
_
,
1
1
1
1
1
.
q
t
( )
1
]
1
1
1